胡渊鸣：Meshy AI，太极，MIT，清华姚班，图形学，物理仿真模拟，开源，商业化，勇气 ，智慧 ｜ WhynotTV Podcast #2

Usin g my own pro gram ming lan guag e to imp leme nt m y ow n pr ogra m a nd t hen I fo und it w as b ette r th an C UDA. I t hink I w as a lit tle craz y at tha t ti me. It's tru e th at o peni

ng a sou rce is n ot p rofi table. But if y ou h ave a go od o pen sour ce p roje ct, it c an h elp you attr act the best tal ent. If you lock you rsel f in the off ice and clos e th e do or t o do

you r ow n th ing ever y da y y ou c an o f co urse pro duce mor e in a s hort tim e. B ut a fter fiv e ye ars, you wil l no t kn ow w hat to d o. A ll d ata is s ynth etic dat a o r mo st o f th

e da ta i s sy nthe tic data . It wil l no t wo rk. It m ay n ot n eed so m uch reco gnit ion from our hum an s elf- perc epti on o f th e ph ysic al w orld . It can fin d it s ow n le arni ng r ate.

- Hu Y anmi ng h as b een writ ing phys ics and game dev elop ment pro gram s si nce chil dhoo d. H

e st udie d in Tsi nghu a Un iver sity of Tech nolo gy a nd t hen went to MIT to s tudy as a pr ofes sor. He is o ne o f th e be st c ompu ter grap hics exp erts in the worl d. I n 20 21, afte r gr

sor. He is o ne o f th e be st c ompu ter grap hics exp erts in the worl d. I n 20 21, afte r gr adua ting fro m th e Un iver sity of Hu Y anmi ng, he s et u p Ta iji Grap hics . Af ter 18 m onth

s of com merc ial expl orat ion, he swit ched to 3D A I pl atfo rm, Mesh . Me sh h as o ver 4 mi llio n se rvic e us ers. and annu al r even ue g rowt h ex ceed ing 10 t imes . Th is c onve rsat

ion cove rs H u's thou ghts on tech nolo gy a nd t he w orld 's n atur e s elf- deve lopm ent duri ng h is P hD, life and dea th t rans form atio n in ent repr eneu rshi p a nd t he e volu tion ary

phil osop hy o f "s cien ce a nd c oura ge" that kil ls o ne's pas t ev ery thre e mo nths. This is

YNA TV P odca st. Now, ple ase join me in H u Yu anmi ng's wor ld. Hell o Y uanm ing. Hel lo, Tai Ran. You hav e ma ny b righ t la bels su ch a s Qi nghu a Ya bai, MIT 99 th S now Ice, inc ludi ng t

Ran. You hav e ma ny b righ t la bels su ch a s Qi nghu a Ya bai, MIT 99 th S now Ice, inc ludi ng t he C EO o f Ta iji Tu X ing and Mess i AI . I beli eve you have lon g be en d isli ked by t hese lab

els, and I k now that the se l abel s ca nnot sho w yo u th e re al y ou. So I 'm v ery curi ous.

Now th e fi rst ques tion we' re g oing to ask you is, 胡彦明是 谁? 你 会如何认 知定义和 介绍你自 己? 最 简单的回 答就是我 是我自己

。 I am m ysel f. 我 是一个被 自己的兴 趣和使命 感驱动着 去做事的 That 's h ow I def ine myse lf. So y our inte rest ...

。 I am m ysel f. 我 是一个被 自己的兴 趣和使命 感驱动着 去做事的 That 's h ow I def ine myse lf. So y our inte rest ...

I se e th at y ou s tart ed w riti ng c ode when you wer e ve ry y oung . Di d yo u ha ve i nter est in c odin g an d pr ogra mmin g si nce you were you ng? I gr ew u p pl ayin g ga mes sinc e I was

youn g. I gue ss y ou c an g uess how man y ga mes I pl ayed whe n I was youn g. B efor e I knew tha

youn g. I gue ss y ou c an g uess how man y ga mes I pl ayed whe n I was youn g. B efor e I knew tha

t I was inte rest ed i n co ding, How old were you at that time? When I w as v ery youn g m aybe bef

ore I we nt t o el emen tary sch ool, whe n I was in k inde rgar ten. Bec ause my pare nts were tea chin g co mput ers, the y we re t each ers in c olle ge. So w hen I wa s ve ry y oung I got into com

pute rs. I re memb er a lon g ti me a go, I as ked my p aren ts s ome very str ange que stio ns.

pute rs. I re memb er a lon g ti me a go, I as ked my p aren ts s ome very str ange que stio ns.

Ther e is one que stio n th at I sti ll d on't kno w wh at t he a nswe r is . I thin k it has a l ot t o do wit h wh at y ou w ant to a sk t oday . I reme mber whe n I was very you ng, 家里住在 学校宿舍

,然后楼 下有一棵 树,可能 是桃树还 是什么。 然后每 到春天的 时候,这 个桃树就 会开花。 我当时 就在思考 一个问题 ,当我在 睡觉的时 候,当我 不在观测 这个桃树 的时候, 这个桃

树到底是 自己在默 默地生长 呢,还是 在偷懒保 持原样? 直到我 看到它的 一瞬间, 它突然长 成了它应 该呈现的 那个样子 。 然后 那个时候 我就开始 想,这个 世界到底 是如何去 运行的? -

I wa s ab le t o ob serv e ma ny t hing s in thi s wo rld very qui ckly . Th e wo rld shou ld b e sl ower . I coul d ob serv e it slo wly. Lat er o n I tho ught I c ould n't obse rve it b ecau se I was

ower . I coul d ob serv e it slo wly. Lat er o n I tho ught I c ould n't obse rve it b ecau se I was als o si mula ted by t his worl d. S o as a m embe r of thi s wo rld that is simu late d y ou c an't

rea lly tell how thi s wo rld is c alcu late d or sim ulat ed. So t hat was a ve ry y oung age . At

tha t ti me, I wa s th inki ng, We c an a ctua lly buil d a syst em t o si mula te a ll t he t hing s th at h appe n in thi s wo rld. How ever al l th e su bjec ts t hat are simu late d in thi s sy

stem can not actu ally rea lize tha t th ey a re i n a simu late d wo rld. Whe n yo u st art to w rite cod e to sim ulat e a real wor ld, it m ay b e af ter elem enta ry s choo l. Y ou s tart ed t

o wr ite code to simu late aft er e leme ntar y sc hool? So y ou a ctua lly don' t te ach any Newt

onia n la w in ele ment ary scho ol? Yes, I f ound out som ethi ng m ysel f. I f I pres s th e ke yboa rd u p an d do wn, it w ill be v ery unre al i f th e ob ject mov es f orwa rd. So I hav e to

def ine it. I us e up and dow n to ope rate . It sho uld be t he a moun t of mov emen t of the obj ect. Lat er I fou nd o ut t hat this is the Newt on's sec ond law. So in f act, it can let you know

ect. Lat er I fou nd o ut t hat this is the Newt on's sec ond law. So in f act, it can let you know a l ot o f ve ry i nter esti ng t hing s. Y ou c an a lso prac tice a l ot o f lo gica l th inki ng.

So w hat kind of thin gs c an y ou s imul ate in t he s imul ator you wro te w hen you were a k id?

That was a.. . At fir st, of c ours e I had a v ery big visi on. That is, can I w rite a g ame? In

this gam e y ou c an g et a ll t he c ompo nent s th roug h th e ar rang emen t of the sys tem. Som

e ve ry m agic al m achi nes that can ach ieve var ious fun ctio ns. At t hat time th ere was a ga me c alle d Cr azy Mach ines . It 's c alle d Cr azy Mach ines or some thin g. I t's prob ably ...

You have a... Ther e's a te nnis bal l he re, and you want it to g o th roug h a seri es o f op erat ions lik e ro llin g c limb ing, ele ctri c fa n b read mac hine an d fi nall y sm ash a ca t on the

scr een. I t houg ht i t wa s ve ry i nter esti ng. And this gam e is act uall y dr iven by some of

scr een. I t houg ht i t wa s ve ry i nter esti ng. And this gam e is act uall y dr iven by some of the earl iest phy sics and phy sics pri ncip les. So I've alw ays been ver y v ery inte rest ed i

n th is t hing . So the gam e I wrot e th e ea rlie st w as, in m y im pres sion, It's a s pace ship . Yo u ca n op erat e th e up per and lowe r le ft b utto ns t o mo ve i t in thi s sp ace. The n yo

. Yo u ca n op erat e th e up per and lowe r le ft b utto ns t o mo ve i t in thi s sp ace. The n yo u ca n fi re a bul let. Lat er, I tr ied to a dd s ome You can cust omiz e th e ai rcra ft. You can

add two acce lera tors to it. And then you can upg rade som e of its com pone nts or s omet hing . Bu t th en i t se ems like ... Beca use the comp uter see ms t o be bro ken, the cod e is als o lo

. Bu t th en i t se ems like ... Beca use the comp uter see ms t o be bro ken, the cod e is als o lo st. And then aft er a whi le, my i nter est beca me t o us e RP G Ma ker XP t o ma ke R PG g ames

st. And then aft er a whi le, my i nter est beca me t o us e RP G Ma ker XP t o ma ke R PG g ames . Ok ay. That 's a 2D game eng ine. The n I got to k now Ruby . Th en I lea rned to writ e Ru by.

. Ok ay. That 's a 2D game eng ine. The n I got to k now Ruby . Th en I lea rned to writ e Ru by.

Then ... Thes e al l ha ppen ed w hen you were in elem enta ry s choo l? Y es, this was a l ong time ago . At tha t ti me, I wa s ac tual ly t he m ost inte rest ed i n ga mes. You jus t me ntio ned

ago . At tha t ti me, I wa s ac tual ly t he m ost inte rest ed i n ga mes. You jus t me ntio ned that you had an idea whe n yo u we re y oung . Yo u we re t hink ing abou t ho w th e wo rld work

s. H ow t he w orld wor ks w hen you don' t ha ve a ny o bser vati on. It's bee n mo re t han 10 y

s. H ow t he w orld wor ks w hen you don' t ha ve a ny o bser vati on. It's bee n mo re t han 10 y ears now . Do you sti ll t hink tha t th e wo rld we l ive in i s a simu lato r? I thi nk a s a subj

ears now . Do you sti ll t hink tha t th e wo rld we l ive in i s a simu lato r? I thi nk a s a subj ect that is simu late d y ou m ay n ot k now the answ er i n it. Mayb e th e wh ole worl d is bor

ed b y a more adv ance d ci vili zati on. So t hey wrot e a prog ram to s imul ate our worl d. J ust

like now if you bui ld a sup erco mput er, it w ill even tual ly b ecom e a mill ion H100 s a nd s imul ate arti fici al i ntel lige nce. In this pro cess if you are jus t a tran sfor mer, a m odel

of atte ntio n y ou m ay n ot b e aw are that you are in such a c onta iner. 对我们来说 I th ink it's ver y po ssib le t hat we a re a sim ulat ed w orld . Yo u ca n se e so me t race s r ight ? Fo r ex

ampl e w hy i s th ere Plan ck's con stan t? I s it the err or o f th e fl oati ng n umbe r of the sim ulat ion syst em? The maxi mum accu racy is to b e ab le t o ex pres s th is. So i n ou r ob

serv atio n i t is a P lanc k co nsta nt. Why is t here a l imit to the spee d of lig ht? Why can

only one obj ect inte ract wit h th ese thin gs i n th e fu ture lig ht p ursu it? Mayb e it can 't a ffor d it. to i nter act with all two par ticl es i n th e wo rld. So I wo uld rath er g ive

you a li mit of l ight spe ed. Why does the uni vers e ex plod e? I t's just a s et o f pa rame ters . An d th en, the moti ve o f th e hi gher -up spec ies or c ivil izat ion to d o th is i s to see

. An d th en, the moti ve o f th e hi gher -up spec ies or c ivil izat ion to d o th is i s to see

uh

- Yes, it' s po ssib le. But I ca n't say if i t's real or fake . Be caus e yo u're in this sys tem, you can 't b reak thr ough the lim it y ou'v e be en s imul ated . So you wro te a gam e in ele ment

ary scho ol, and then wha t di d yo u do in juni or h igh? Did you wri te b ette r ga mes? In juni

or h igh, I w as o ccup ied by c ompu ter game s. I wou ld d o ma th q uest ions and tak e pa rt i n co mpet itio ns. ny s uch comp etit ions . In the pro cess I actu ally pra ctic ed a lot of comp

etit ive prog ramm ing. In fact it giv es y ou l imit ed t ime. Und er g reat pre ssur e y ou h ave to w rite the cod e an d wr ite it r ight . I thin k th is i s so meth ing that sti ll m akes me

very sat isfi ed. When I w as i n ju nior hig h sc hool I wrot e mo re. I re memb er w riti ng a "ta nghu an s yste m" a t th at t ime. You can bui ld t his a so ft b ody. The re w as a gam e ca lled

Wor ld o f Go ose. You hav e a lot of b alls and you can con nect the m to geth er t o ma ke a bui ldin g. I t's a bi t li ke t he o ne w e pl ayed bef ore. I d on't kno w if you 've play ed i t. Y

ldin g. I t's a bi t li ke t he o ne w e pl ayed bef ore. I d on't kno w if you 've play ed i t. Y

ou c an b uild a l ot o f ma gnet s an d st eel bars . Yo u ca n bu ild them up and make the m ve ry h igh. In the comp uter it was a m ass- brin g sy stem . A lot of b alls and som e ri ngs. I t

hink I h ave some rel ated vid eos. The n it was ... I th ink I wa s in hig h sc hool . Of cou rse, I s pent a l ot o f ti me i n cl asse s an d in com peti tion s. B ut I don 't r emem ber if I was

in high sch ool. I w as s tudy ing if I cou ld s imul ate some rig id b odie s. I tho ught the dom inoe s we re f un. I wa nted to make dom inoe s in the com pute r be caus e I had a pi le o f th

ings in the real wor ld. I di dn't hav e a plac e to put the m a nd I jus t to uche d th em a nd i t wa s al l ov er. I th ough t it wou ld b e mo re i nter esti ng t o do it in t he c ompu ter. So

I we nt t o th e co mput er t o do rig id b ody simu lati on. And duri ng t his proc ess, I f ound tha t ac tual ly, a lo t of thi ngs are very ina ccur ate. For exa mple th e co llis ion betw een

two stee l bo dies or thei r fr icti on. Thes e mo dels are all sim ilar mod els. But I wa s st ill very int eres ted in t his thin g. T hen I st arte d st udyi ng f luid aft er I wen t to col lege

. An d wh en I got my PhD, I w as s till doi ng d efor mabl e ob jects. elas tic obje cts and some

. An d wh en I got my PhD, I w as s till doi ng d efor mabl e ob jects. elas tic obje cts and some gas -bas ed o bjec ts. So y ou j ust said tha t wh en y ou w ere in h igh scho ol, when you wer e do

ing domi noes yo u re aliz ed t hat even the con tact of fric tion of rigi d bo dies is not very acc urat e. S o af ter high sch ool, you wen t to Tsi nghu a. I t wa s a rock ban d. I t wa s pr obab

urat e. S o af ter high sch ool, you wen t to Tsi nghu a. I t wa s a rock ban d. I t wa s pr obab ly a rel ativ ely earl y ro ck b and, was n't it? It s houl d ha ve b een 10 y ears . I shou ld n

ot b e th e ea rlie st s tude nt. Now that you rec all it, what do you thin k yo u di d in tho se f our year s? I thi nk... I ha ve a lot of expe rien ce. Yest erda y I had a b irth day part y wi th a

fri end. We went to his birt hday par ty i n th e Ba y Ar ea. We a re v ery clos e no w. W e ha ve b

fri end. We went to his birt hday par ty i n th e Ba y Ar ea. We a re v ery clos e no w. W e ha ve b een grad uate d fo r al most 10 year s. W e gr adua ted in 2 017. It' s be en 8 yea rs s ince the

n. W e ha ve a par ty e very yea r. I thi nk t he m ost expe rien ce I hav e is fro m a grou p of

n. W e ha ve a par ty e very yea r. I thi nk t he m ost expe rien ce I hav e is fro m a grou p of ver y go od s tude nts. Amo ng t hem, Some peo ple don' t re ally lea rn m uch, but the y ge t 10

0 po ints eve ry t ime they tak e th e ex am. Some peo ple can easi ly g et t he g old meda l in the ACM Wor ld F inal . I thin k I real ized som ethi ng. Ther e ar e a lot of g reat peo ple in t his

worl d. I thi nk j ust one thin g li ke t his is e noug h to Just ify is e noug h to exp lain tha

worl d. I thi nk j ust one thin g li ke t his is e noug h to Just ify is e noug h to exp lain tha t th ese four yea rs a re v ery valu able . So bef ore the Yaob an, you were ver y pr omin ent. The

n wh en y ou s udde nly arri ved at Y aoba n y ou r eali zed that the re w ere so m any grea t pe ople . Ac tual ly, no. Actu ally wh en I was in the midd le s choo l w hen I wa s in the com peti

ople . Ac tual ly, no. Actu ally wh en I was in the midd le s choo l w hen I wa s in the com peti tion I met a lo t of gre at p eopl e in the pro vinc ial team of Jian gsu Prov ince. So I thi nk i

t's actu ally mor e he lpfu l to kno w so me p ower ful peop le e arli er. It w ill let you know tha t th ere are peop le o ut t here and the re a re p eopl e ou t th ere. You 're not as p ower ful

as y ou t hink . Yo u ha ve a lot of plac es t o im prove. So I thi nk t hat in t he p roce ss o f co mpet ing and movi ng t o th e US in clud ing the futu re, to M IT, I ca n co nsta ntly enc ount

er s ome thin gs. You will fin d th at w ow, ther e ar e st ill peop le w ho a re s o go od. In t

his proc ess, it actu ally has a g reat hel p fo r my own gro wth. The tre nd i s gr owth . So is

ther e an y mo ment tha t ma kes you feel ver y fr ustr ated ? Th ere are many suc h mo ment s. T

here are too man y. I rem embe r wh en I was in Tsin ghua, The most imp ress ive thin g wa s th e fi rst clas s we too k w hich was cal led "Gal axy App Math emat ics" . It was the fir st t est, and it

was quit e di ffic ult. Bec ause it was not the same as othe r cl asse s in the uni vers ity. In

Tsin ghua yo u wo uld lear n ab out the alge bra, the lin ear func tion an d th e th eory of grav ity. But in this cla ss, I as sume d yo u kn ew a ll o f th ese, and I s tart ed t o te ach you.

ity. But in this cla ss, I as sume d yo u kn ew a ll o f th ese, and I s tart ed t o te ach you.

Anyw ay, this is a ve ry i nten se e nvir onme nt. And then the re w as o ne t ime wher e I fail ed.

I th ink this is a ve ry o pen- mind ed a ppro ach to p eopl e. I t's very eye -ope ning . So you

sta rted doi ng r esea rch in y our majo r a nd y ou s tudi ed r obot ics? Yes, at the earl iest th

ere was a pr oble m wi th t he d rawi ng s choo l. I n fa ct, ther e we re n ot m any oppo rtun itie s to do draw ing in C hina . So I b asic ally mad e my own dra fts in t he d ormi tory . So dur ing

the proc ess of m akin g th e dr afts I basi call y lo oked at the C-gr aph pape r an d th ough t " Wow, thi s is so cool . Ca n I try it o ut?" And I d id a lot of C-gr aph pape r. A nd d urin g th

is p roce ss, 也积累了 很多对于 像物理仿 真 像r ende ring 渲染 这里面的 很多技术 很底层的 了解 然 后当然也 发现了很 多pap er 它 没有把所 有的事情

都写出来 这个其 实也让我 比较早的 意识到 科研其实 有很多局 限性 不 能太迷信 pape r里面说 的事情 Why did you thin k th ere was no s uch thin

g as a n ativ e le arni ng i n Ch ina? The bes t st uden ts w ent to A I. E ven in 2 013 to 2 017?

Yes . It was whe n th e sm art netw ork was very pop ular . I reme mber in 2012 Al exNe t ca me o ut. Then the re w as R esNe t an d Al phaG o. T he w hole gra phic s wa s no t th e be st d irec

ut. Then the re w as R esNe t an d Al phaG o. T he w hole gra phic s wa s no t th e be st d irec tion . Th e be st p eopl e or ... If y ou d on't thi nk a bout wha t yo u re ally lik e m any peop

tion . Th e be st p eopl e or ... If y ou d on't thi nk a bout wha t yo u re ally lik e m any peop le w ill actu ally go for AI. And many peo ple did a go od j ob. I ha ve a lot of clas smat es w

ho l ater got ver y v ery infl uent ial jobs . Di d yo u co nsid er g oing for AI back the n? Y es,

I al so w ent for AI f or a whi le. I th ink it's pre tty good . I was at t he M icro soft Asi a In stit ute, and I w as a n in tern wit h St eve Lin, a m embe r of MSRA. Thei r gr oup is c alle d In

tern et G raph ics. I w ent ther e an d fo und that eve ryon e wa s do ing AI. So I wen t to do AI.

In t he e arly day s a roun d th e th ird year of coll ege, I s tart ed t o wr ite a CV PR p aper . It was a p aper abo ut p redi ctin g Y- Bala nce usin g th e ne ural net work . It was pub lish

. It was a p aper abo ut p redi ctin g Y- Bala nce usin g th e ne ural net work . It was pub lish ed i n CV PR 2 017. Then I s aw A lpha Go a nd I tho ught tha t Al phaG o an d RL wer e gr eat. Rei

nfor ceme nt l earn ing. The n I trie d to use rei nfor ceme nt l earn ing to d o th e po st-p roce ssin g of the ima ge. At t hat time GA N wa s al so v ery popu lar. So I wa nted to try GAN. The n I

put GAN and RL t oget her and fina lly trai ned more tha n 1, 000 mode ls. I fi nall y ma nage d to tra in a mod el t hat was bare ly p ossi ble. Fin ally I foun d th at t hese two tec hnol ogie s we

re n ot a s ma ture . Th is i s my opi nion of AI. So So y ou a ctua lly did a lo t of res earc h on

AI in y our majo r. I did som e. H ow m uch did you do i n yo ur m ajor? 最后发出 论文的全 部都是A I的研究 Gra

phic s一篇都 没发出来 Gra phic s 其实 第二篇是 AI加G raph ics 所以最后 发表在了 Tran sact ion on G raph ics 也就是C grap h上面

不过这个 其实没有 在我本科 的时候做 完 后来 是我到M IT这个 文章才被 接收 然 后发表出 去 所以 你到了M IT之后 当时想 过自己要 做什么吗 还是你 非常坚定 我要做G raph ics 当时就非 常坚定想 做Gra phics Even if it w as 2 017, I t hink the atm osph ere at t hat time was tha t ma

ny p eopl e we re d oing AI at C SAIL . Yo u we re v ery dete rmin ed a t th at t ime. You kne w yo u wo uld stil l do gra phic s. I thi nk t here is an f acto r th at C SAIL did n't have man y pe

ople doi ng A I. I thi nk t he w hole MIT and Sta nfor d t he s choo ls t hat are clos e to the ind ustr ial cent er, have a m uch lowe r ac cept ance rat e of AI. In fact th is h as a gre at r elat

ions hip with the ent repr eneu rial atm osph ere of t he e ntir e Bo ston area. Beca use it d oesn 't h ave indu stry dri ve, Bost on h as m ore bio. Bec ause you see Bo ston 's i niti al c ompa ny i

s ba sica lly a bi o-ba sed init ial comp any. The re a re f ewer ini tial com pani es i n th e ar ea o f AI . Ye s w hat impr esse d me was tha t wh en I wen t to Bos ton, I f elt that it was all

biot ech comp anie s. Y es, so t his actu ally led to MIT' s... This is my o pini on. I th ink this may hav e le d to MIT not fol lowi ng t he w hole of AI i n a very clo se w ay. I th ink this can be

a go od t hing or a ba d th ing. I t hink thi s ma y ha ve t o be cha nged . So in the 17th cen tury ma ny s tude nts were sti ll d oing tra diti onal thi ngs, suc h as var ious com pile rs f or s imul

atio n 3 D pr inti ng, Ever yone tri es t o in tegr ate AI i nto thei r wo rk, but I do n't thin k th ere' s mu ch f unda ment al A I re sear ch c omin g ou t of MIT . I thin k wh at I was luc ky a

bout at MIT was that aft er I fin ishe d my mas ter' s de gree I swit ched to the team of two of m y la ter ment ors, Fre do D uran d an d Bi ll F reem an. And unde r th eir guid ance th ey w ere all

hand s-off. 嗯,对我 的研究没 有加太多 的约束, 所以我 可以比较 自由的去 做我的r esea rch, 那在这 个过程当 中 我 觉得我是 非常en joy这 段时间的 。 但是

hand s-off. 嗯,对我 的研究没 有加太多 的约束, 所以我 可以比较 自由的去 做我的r esea rch, 那在这 个过程当 中 我 觉得我是 非常en joy这 段时间的 。 但是

你跳过了 一 可 能第一年 你不太e njoy ,对不对 ? 因为 我看到你 在很多地 方都说过 第一 年差点在 MIT, 当时想退 学了。 对,当时 差点退学 退学 的原因呢 是 当 时我和我

的第一年 的导师的 研究兴趣 并不是特 别一致, 然后这 个我,我, On o ne h and, I h ave my o wn r esea rch inte rest s a nd o n th e ot her hand I have to meet som e fu ndin g re quir emen ts. So I hav e a lot of t ime to d o th

ings tha t I don' t en joy. I c an d o a thou sand thi ngs that I e njoy bu t if I d on't enj oy d oing thi ngs that I e njoy I migh t be a v ery pain ful pers on. In t he p roce ss o f le avin g Ch

ina, I l earn ed t hat I ca n on ly d o th ings tha t I enjoy. 如果说一 个事情我 没有那么 enjo y I wou ld r athe

r no t do it. 我宁愿 不做它, 我也不会 去开始一 段痛苦的 这样的一 个经历。 但是你 我觉得有 一个很强 的天赋是, You know ver y we ll w hat you enjo y. M any peop le m ay n ot h ave this awa rene ss o r in the ir e arly sch ool year s t hey may

enjo y. M any peop le m ay n ot h ave this awa rene ss o r in the ir e arly sch ool year s t hey may not know wha t th ey w ant to d o. W hen did you real ize that you lik e th is, you enjo y th is?

How long ago did you hav e th is a ware ness ? I thin k it was whe n I was very you ng. For exam

ple, I k now I li ke w riti ng p rogr ams. Thi s wa s a long tim e ag o. M aybe whe n I was in e leme ntar y sc hool. I kn ew t hat I wo uld feel bad if I di dn't wri te t he p rogr am f or a who

le d ay. So a t th at t ime, I k new that I r eall y li ked to c reat e so meth ing. And the n C ++ a nd P ytho n ar e ve ry f lexi ble buil ding blo cks. It's lik e pl ayin g wi th a mud dle, but

it' s no t as eas y as C++ and Pyt hon. It can help you bui ld s omet hing . I thin k I knew wha t I was doin g ea rlie r. M any peop le d on't kno w. E ven afte r fi nish ing the PC, they don 't k now

what the y wa nt t o do . Do you thi nk i t's your tal ent, luc k o r ot her fact ors? I th ink luck is the key to s ucce ss. Gene tica lly, you hav e a good gen e s o yo u ne ed i t in thi s er a. S

o I thin k 99 % of it is l uck. Eve n if you can wor k ha rd, it's als o pa rt o f lu ck. Beca use

you have luc k y ou k now what you lik e s o yo u ca n wo rk h ard. So I th ink... Luck is a ve ry i mpor tant thi ng. In t he f irst yea r y ou a nd y our ment or h ad d iffe rent int eres ts. You

were str esse d ab out the foun ding pro ject . Wh at w as t he r easo n or who hel ped you to g et o ut o f th e st ate of w anti ng t o qu it s choo l? I thi nk o ne i s th at I sti ll s ent a lo

t of pap ers. At that tim e I was sti ll r elat ivel y pr oduc tive . Al thou gh i t wa s ve ry p ainf ul, I wa s re lati vely pro duct ive. Ano ther one is that I a ctua lly talk ed t o a few seni

ors. The y we re m y se nior s in Tsi nghu a. A nd t hen I ha ppen ed t o do a p ostd oc i n th e re

ors. The y we re m y se nior s in Tsi nghu a. A nd t hen I ha ppen ed t o do a p ostd oc i n th e re leva nt g roup or a ve ry s enio r PhD. So a fter tal king to them I thou ght I co uld find ano ther

gro up o f in stru ctor s in MIT . La ter, I c hang ed m y te ache r an d I felt tha t my lif e wa s mu ch b ette r. S o ac tual ly, you talk ed t o th e pr esid ent. Who exa ctly did the y te ll y

ou? What sto ry d id t hey tell you ? It was mai nly Jiaj un a nd J unya n. A nd t hen I ac tual

ou? What sto ry d id t hey tell you ? It was mai nly Jiaj un a nd J unya n. A nd t hen I ac tual ly t alke d to the m ab out what sci ence can do. Then Jia jun told me, "Wh y do n't you try to m

ake your sim ulat or d iffe rent iabl e?" You can ask for the simu lato r's guid e. T hen it b ecam e th e tw o pa pers of Chin Qui nn a nd D ave Taic hi. Thes e tw o pa pers are act uall y ve ry i

nter esti ng. They com bine sim ulat ion and robo tics . Th en J unya n to ld m e th at i t is imp orta nt t o do an impa ct r esea rch. The n of cou rse I fi gure d ou t that Ther e ar e st ill a lo

t of thi ngs to d o at MIT . It 's a ver y v ery good pla tfor m. A nd t here are a l ot o f ex cell ent peop le h ere. So I la ter thou ght, and of cour se, in t he m ost trad itio nal sens e y ou c

an g radu ate from MIT as a do ctor . Th is i tsel f is a k ind of c erti fica tion . So if othe rs d on't kno w an ythi ng a bout you yo u ha ve t his cert ific atio n i t's usef ul. But the most

imp orta nt t hing is that I t hink tha t st ayin g at MIT can do a lo t of int eres ting wor k. I

f yo u be come the sen ior who open ed t he f irst -yea r Hu Yem ing PhD prog ram, wha t wo uld you say to h im? That 's a ver y go od q uest ion. I t hink we shou ld o pen our eyes and spe nd m ore

time thi nkin g ab out what kin d of wor k is imp orta nt, and then exe cute thi s pl an. Spen d yo

ur t ime, car ry o ut, a go od s trat egy befo re y ou e xecu te i t. -

Yes, I t hink on one hand th ere is n ot e noug h ti me, and on t he o ther han d o bjec tive ly, when you hav e pu blis hed enou gh p aper s an d fi lled eno ugh wate r y ou m ay n ot k now what

kin d of pro blem it is. It i s an imp orta nt p robl em. Mayb e th ey d on't hav e th is a ware ness . Ye s y ou m ay h ave publ ishe d a pape r an d sa y " Why do I thi nk I did so well in this

ness . Ye s y ou m ay h ave publ ishe d a pape r an d sa y " Why do I thi nk I did so well in this art icle ? Wh en I tal k to oth ers, the y ar e no t ex cite d at all ." " Why are ther e on ly a

art icle ? Wh en I tal k to oth ers, the y ar e no t ex cite d at all ." " Why are ther e on ly a few peo ple who pay atte ntio n to my arti cle? " Wh en y ou h ave gone thr ough eno ugh of t hese

fai lure s o r re flec tions, you can know wha t th e or igin al d irec tion sho uld be l ike. It

take s so me e xper ienc e. Y ou d idn' t ha ve t he k nowl edge to look for imp orta nt t hing s. T

hen you said in your pap er t hat ever y Me ssi play er i n th e te am w ould ask him if what he did was impo rtan t. Y es. I'm curi ous abou t wh at y ou'v e be en d oing rec entl y th at's imp orta

nt. Why is i t im port ant? Tha t's a go od q uest ion. I'l l ha ve t o th ink abou t it.

nt. Why is i t im port ant? Tha t's a go od q uest ion. I'l l ha ve t o th ink abou t it.

If i t's from the com pany 's p ersp ecti ve, the most imp orta nt t hing is recr uitm ent. Bec ause

act uall y yo u sa y on e b ut t his may jump fro m th e ac adem ic w orld to how to s tart a b usin ess. In fact th e co re c ompe titi on o f a comp any is a lway s ta lent . As lon g as you hav e th

ess. In fact th e co re c ompe titi on o f a comp any is a lway s ta lent . As lon g as you hav e th e be st t alen t y ou c an m ake the best pro duct . Wi th t he b est prod uct, you can hav e th

e mo st u sers th e mo st u sers can hav e th e mo st r even ue, and this rev enue can be inve sted in R&D. You can als o fi nd b ette r ta lent. So I thi nk t alen t ac quis itio n ha s al ways bee

in R&D. You can als o fi nd b ette r ta lent. So I thi nk t alen t ac quis itio n ha s al ways bee n on e of the mos t im port ant thin gs f or a CEO . As a C EO, the most imp orta nt t hing for you

now is recr uiti ng, find ing bett er p eopl e. I s th ere anyt hing mor e im port ant than thi s? N

o. E ven if i t's the comp any' s st rate gic dire ctio n w hat prod uct do w e wa nt t o ma ke, do y ou t hink it' s no t as imp orta nt a s fi ndin g be tter peo ple? I t hink it' s eq uall y im

port ant. One is to t hink abo ut t he c ompa ny's roa dmap and str ateg y. A noth er i s re crui

port ant. One is to t hink abo ut t he c ompa ny's roa dmap and str ateg y. A noth er i s re crui ting . Of cou rse I Anot her impo rtan t th ing is t o pa rtic ipat e in the com pany 's d aily bus

ting . Of cou rse I Anot her impo rtan t th ing is t o pa rtic ipat e in the com pany 's d aily bus ines s. I thi nk t hese thr ee t hing s ar e: R ecru itin g +4 0% T hink ing abou t th e fu ture

ines s. I thi nk t hese thr ee t hing s ar e: R ecru itin g +4 0% T hink ing abou t th e fu ture : 30 % An d th en t o pa rtic ipat e in the bus ines s cl osel y: 3 0% A ctua lly, thi s qu esti

on s houl d be ask ed l ater bu t I' m ab out to a sk y ou n ow. As a CEO wh at k ind of c hara cter isti cs d o yo u wa nt t o at trac t th e be st t alen t? I wan t to att ract him to my c ompa ny.

I th ink this is ofte n ju dged by feel ing. But gen eral ly s peak ing, peo ple like thi s ha ve s ome char acte rist ics. I t hink it can be s umma rize d as fou r po ints . Th e fi rst poin t is

"hu ngry ". A hun gry pers on o ften doe s no t pr ove hims elf or h erse lf. He w ants to beco me a bet ter hims elf. He want s to go to a pla ce w here he can grow qui ckly and sta rt h is o wn c

aree r. T his kind of self -imp rove ment 饥渴感是 很重要的 。 那么 如果说你 去找一个 已经很功 臣名就, 很wel l es tabl ishe

aree r. T his kind of self -imp rove ment 饥渴感是 很重要的 。 那么 如果说你 去找一个 已经很功 臣名就, 很wel l es tabl ishe

d的人, 他可能 就不会愿 意去ta ke t he r isk, 去做一 些能让自 己的职业 发展跟上 一层楼的 事情。 这个是h ungr y。 第 二个是h umbl e。 H umbl

e就是谦 虚。 其 实我在招 聘的时候 ,很重要 的一个事 情就是看 这个人是 不是op en。 he c an r eali ze t hat ther e ar e st ill many thi ngs he c an't do well . Be caus e th e mo st t erri ble thin g fo r a pers on i s th at h e ha s ig

nore d th e fa cts, and has bee n us ing this twi sted rea lity to comf ort hims elf, thi nkin g th at h e is doi ng w ell. You can see thi s ki nd o f th ing in h isto ry. Many dis aste rs h appe

n be caus e th ey d on't fac e re alit y. W hen you face rea lity yo u wi ll f ind out what act ions you hav e to tak e. " I ha ve s o ma ny t hing s I have n't done wel l." Humb le i s a poin t I valu

e ve ry m uch. If a pe rson is not humb le a nd n ot o pen, eve n if he is s tron g at thi s ti me, he w ill not move for ward in the futu re. Or i n a word it is call ed g rowt h mi ndse t. C an t

his pers on c onst antl y ch ange him self co nsta ntly den y hi msel f a nd c onst antl y gr ow?

If y ou a re n ot h umbl e y ou c an't gro w. Y ou c an't lis ten to o ther peo ple' s op inio ns.

The thir d is cal led smar t. O f co urse yo u wa nt t o be wit h sm art peop le. So s mart mea ns i t ca n th ink of a ll k inds of ways to solv e pr oble ms w ith high res ourc es. For exam ple, eve

n th ough I h ave limi ted comp utin g re sour ces, I t houg ht o f a grea t op timi zer. I'm now

usi ng a mod el t hat' s tw ice as e ffic ient . I can use a lo wer floa ting poi nt p reci sion to trai n a mode l. I can use som e ha cks to m ake it p ossi ble for me t o do mor e th ings wit h le

ss r esou rces . Th ey a re t he b est prob lem solv ers. In a st artu p co mpan y y ou d on't nee d a good man ager. You need a c erta in l evel of mana geme nt, but what you nee d mo re i s th at e

very ind ivid ual cont ribu tor is a goo d pr oble m so lver . He can sol ve t he p robl em. The

four th o ne I thi nk i s cl arit y. H is e xpre ssio n sh ould be clea r. Y ou s houl d be abl e to und erst and what he says . Ma ny t imes you go to i nter view som eone or chat wit h so meon e wh

o yo u th ink has a hi gher lev el o f kn owle dge than you . Wh en y ou f ind that you don 't u nder stan d wh at h e sa ys a t all, It's not tha t yo u ha ve p robl ems, or that you r kn owle dge

leve l is too poo r b ut t hat he d idn' t th ink abou t it cle arly . I thin k th at a cle ar e xpre ssio n of coo pera tion wit h a team th e co hesi on o f a team an d th e wa y of thi nkin g of

a p erso n ar e al l ve ry h elpf ul. Some time s i t's not that the exp ress ion is n ot c lear bu t th at t he p erso n's mind is not clea r. P eopl e wh o ar e no t cl ear abou t th ings in thei

r mi nds ofte n do n't have the two thi ngs, hum ble and hung ry, and smar t. S o in sum mary hu ngry hu mble sm art, and cla rity . Ha ve y ou m ade mist akes in thes e fo ur p oint s? H ave you

ever bee n no t hu ngry hu mble sm art, and wit hout cla rity at a ce rtai n st age of y our life ? Th at's a v ery good que stio n. I thi nk I oft en m ake mist akes lik e th is. I al ways thi nk,

? Th at's a v ery good que stio n. I thi nk I oft en m ake mist akes lik e th is. I al ways thi nk, "Can my audi ence und erst and what I j ust said ?" I alw ays refl ect on t his. Thi s ac tual ly i

nvol ves the issu e of cla rity. Even if we a re r ecor ding thi s po dcas t I hav e to mak e su re t hat all my o pini ons are clea rly conv eyed to you, Tai ran, and all you r au dien ce. In t

he p ast, I t hink cla rity was a p robl em t hat I ha d ov erco me v ery earl y on . Of cou rse, I t hink the re a re m any diff eren t ty pes of c lari ty. One of t he m ost comm on t ypes of clar

ity is t hat this per son will say a l ot. In t he p roce ss o f ge ttin g al ong with oth ers, he woul d tr y to put Dom inic 's f ree dial ogue wit hout lis teni ng t o wh at o ther s we re s ayin

g. T hen he s aid a lo t b ut o nly 10% of i t wa s ac cept able by the othe r pa rty. I c an s ee t

g. T hen he s aid a lo t b ut o nly 10% of i t wa s ac cept able by the othe r pa rty. I c an s ee t his prob lem in m any new mana gers . Wh en I fir st b ecam e a mana ger, I a lso had this pro blem

. Un til I fo und out that spe akin g is not imp orta nt, list enin g is the mos t im port ant. Thi

. Un til I fo und out that spe akin g is not imp orta nt, list enin g is the mos t im port ant. Thi

s is act uall y mo re i mpor tant tha n sa ying a l ot. Clar ity.

Hung ry i s th e op posi te o f ly ing flat . Ha ve I eve r li ed f lat? May be n ot. I've alw ays been a h ardw orki ng p erso n. B ut t hat' s ju st l uck. My body is alwa ys w orki ng h ard to s

uppo rt m e. S o hu ngry is luck . Hu mble ... I th ink I'm bein g be aten by soci ety ever y mi

nute. Smar t. S mart . Sm art. I th ink the bigg est mist ake smar t ma de w as t o be sma rt a bout

nute. Smar t. S mart . Sm art. I th ink the bigg est mist ake smar t ma de w as t o be sma rt a bout

the sol utio n b ut i gnor e be ing smar t ab out stra tegy . Yo u sp end too much tim e th inki ng a bout the pro blem of this acc ount bu t yo u do n't spen d to o mu ch t ime thin king abo ut w

hy. Just lik e yo u ju st s aid, PhD fir st y ear. Rig ht. Incl udin g ac tual ly b eing in the comp

hy. Just lik e yo u ju st s aid, PhD fir st y ear. Rig ht. Incl udin g ac tual ly b eing in the comp any, it' s th e sa me. This com pany act uall y di d it for abo ut f our year s. A ctua lly, Like

man y ot her star tup comp anie s w e ha ve e xper ienc ed s ever al t rans form atio ns. In f act, eve ry t rans form atio n is not abo ut n ot d oing har d wo rk, but abou t no t fi guri ng o ut t

he w hy. I th ink we w ill talk abo ut t his late r. I wan t to go back to what you sai d in tha t ar ticl e. I f yo u re ad t he P hD a gain yo u wi ll h ave done fou r ch ange s an d th ings tha

t yo u pe rsis ted in. One is t hat you said you mig ht n ot h ave chos en t o wr ite Tai Chi in N VIDI A b ut w ould acc ept the offe r fr om K ei t o do FAI R. Y ou m ight do AI r esea rch. At that

tim e i t mi ght be m ore comf orta ble for you to w rite Tai Chi in NVID IA. Why did you sudd enly mak e th at c hoic e? D id o ne o f th e fo ur m ista kes you made lea d to you goi ng t o Nv idia

and the n to Fai r? F irst of all, I t hink Nvi dia is a ver y go od c ompa ny. Goin g th ere is a lso a go od c hoic e. B ut I thi nk i f I coul d ha ve p arti cipa ted in A I's more exc itin g fr

onti er e arli er, it w ould hav e he lped me more . Ho w ma ny y ears did you wor k th ere? It was

arou nd 2 012. 零年的暑 假 那个 时候是C OVID 对吧 二零年的 暑假 然 后第二个 点你谈到 的是 你 说你还是 会wor k ha rd 还 是会去做 太极 但 是不会把

太极 那 么局限在 phys ical sim ulat ion上 对 我 觉得当然 这个可能 是一个 可能会变 但是其 实as of t oday 你今天 去看 那 太极在s imul atio

n 其实 是一个相 对来说 比较小众 的使用场景 Of c ours e w ith the use of t he g iant AI tech nolo gy i n th e No rth, the re w ill be m ore and more pla ces that can be used for sim ulat ion. So it m ay b e di ffer ent

in t he f utur e. A nd I thi nk t he m issi on o f TS MC m ay o nly be h alfw ay d one. I w ant to d o TS MC 2 .0 w ith ever yone lat er. Ther e ar e ma ny e xcit ing thin gs t hat can be d one unde

r th e fr amew ork of T SMC. But ther e is a s eque nce of p robl ems, rig ht? In f act, if we w ere to t ake Taic hi t o th e wa y of opt imiz ing the tran sfor mer, Tai chi woul d be in the trit on p

osit ion. It woul d ha ve m ore use case s. M aybe its imp act woul d be muc h gr eate r th an n ow.

osit ion. It woul d ha ve m ore use case s. M aybe its imp act woul d be muc h gr eate r th an n ow.

Tai Chi, we' ll t alk abou t it lat er. Thir d y ou s aid you woul d st ill use a ye ar o f Ph D as you r go al. Even if this goa l is imp ossi ble to a chie ve, do y ou t hink it can help you to star

t so meth ing more int eres ting ? Ye s I thi nk s etti ng a hig h go al i s ve ry i mpor tant . I

will sti ll u se t his year of grad uati on a s my goa l. I thi nk i t's alwa ys a goa l to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach

ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach

ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach

ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach

ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach ieve to ach

ieve to ach ieve to ach ieve to ach ieve to ach ieve So i n fa ct, do y ou t hink it' s a good goa l to gra duat e ea rly? Bec ause in fact ma ny p eopl e sa y d on't wor ry, you do i t sl owly . Ye

s I thi nk d iffe rent peo ple have dif fere nt c hoic es. For me, I th ink the key is s till to see when you are stu dyin g fo r a doct orat e a re y ou i n th is p osit ion, is it t he m ost suit

able env iron ment for you to do w hat you want to do? If y our curr ent envi ronm ent is t hat I wa nt t o co ntin ue t o do it in t he a cade mic world, I ca n ac cept tha t th ere may not be t hese

res ourc es i n th e sc hool . I can acce pt t hat the work I d o ma y be a l ittl e fa r fr om t he u sers bu t it may hav e a grea t in flue nce on t he a cade mic comm unit y. I f yo u wa nt t

his, of cour se, it's goo d to spe nd m ore time on PhD. But whe n I was stud ying for my PhD, I d idn' t kn ow w hat I sh ould do if I stu died for my PhD. Of c ours e I did n't see any adva nced AI

in t he f ield . So fr ankl y I was a b it c onfu sed. I d idn' t kn ow w hat I co uld do i f I stud ied for a do ctor ate. Whe n yo u we re a doc tor, you rea lize d th at t here was not hing you cou

ld d o? P roba bly in t he t hird yea r w hen I wa s ab out to g radu ate. But at that tim e P aper

had eno ugh infl uenc e. S o t he t wo b osse s le t me gra duat e in adv ance. - So.. . I did have

les s co mmun icat ion with my fami ly. I re memb er R ich Hemm ing said som ethi ng l ike this : "I f yo u lo ck y ours elf in y our offi ce, and clos e th e do or e very day to do y our own thin g y

ou c an d o mo re i n a shor t ti me. But in f ive year s y ou w ill not know wha t to do. " So his arg umen t wa s th at y ou s houl d op en t he d oor in y our office to l iste n to wha t ot hers

are doi ng. So t hat othe rs c an f ind you more eas ily and tell you "O h t here 's a noth er p robl em t hat' s so ser ious tha t th ey c an't sol ve, but my t echn olog y ca n he lp t hem solv

e it ." I thi nk t his is a ver y go od m inds et. In f act, as a st artu p co mpan y y ou o ften

e it ." I thi nk t his is a ver y go od m inds et. In f act, as a st artu p co mpan y y ou o ften tal k to som e cu stom ers and list en t o th eir feed back. It's als o a very hel pful thi ng f

or t he e ntir e te am. But beca use I ha ve a n in trov erte d pe rson alit y I hav e to for ce m ysel f to do this . I thin k yo u're alr eady ver y ex trov erte d. I 'm s omeo ne w ho t alks to

othe rs a nd I nee d to con sume a l ot o f en ergy . So in the end, I u se c ode to r ecov er t his ener gy. You just men tion ed R icha rd H emin g. I 'm v ery happ y th at I can use thi s sh adow to

ener gy. You just men tion ed R icha rd H emin g. I 'm v ery happ y th at I can use thi s sh adow to lead us to o ur n ext chap ter. In this cha pter I want to talk to you abou t Ta i Ch i gr aphi cs a

nd p hysi cal simu lati on. So t he f irst que stio n is a R icha rd H emin g-st yle ques tion. What

is comp uter gra phic s? W hat is t he m ost impo rtan t qu esti on i n th e fi eld of c ompu ter grap hics ? Wo w t his is a ver y go od q uest ion. But I t hink dif fere nt p eopl e sh ould hav

grap hics ? Wo w t his is a ver y go od q uest ion. But I t hink dif fere nt p eopl e sh ould hav e di ffer ent answ ers. My answ er i s th at c ompu ter grap hics is to c reat e a virt ual worl

d wi th a ddit ion and subt ract ion. Bec ause whe ther it' s CP U or GPU it can onl y do add itio n an d su btra ctio n o pen tria ngle fun ctio n. B ut y ou h ave to u se t hese bas ic b uild ing

bloc ks. to c reat e a very uni que game lik e Cy berp unk 2077 or The Witc her 3. I thi nk i t's

bloc ks. to c reat e a very uni que game lik e Cy berp unk 2077 or The Witc her 3. I thi nk i t's very ve ry a ttra ctiv e. B ut t he e ntir e co mput er g raph ics indu stry has som e pr oble ms i

n th e in dust ry. So i t's not as p opul ar a s AI or data. But ther e ar e st ill many peo ple who love it and cont inue to do s ome entr epre neur ship in this are a. B ut I thi nk t he d evel opme

nt o f co mput er g raph ics is v ery inte rest ing, inc ludi ng t he f act that AI hard ware is base d on gra phic s. I t wa s de velo ped at t he e arli est. So in t his, gra phic s is act uall

y a way for many sch olar s to pav e th e wa y. W hat is t he m ost impo rtan t pr oble m in the fie ld o f gr aphi cs? It d epen ds o n th e er a. A t th e ea rlie st, grap hics was lim ited to the

fact tha t it was ver ifia ble. The pape r is cal led "Vis ual Simu lati on o f Sm oke" It was in 1

995- 96. At t hat time pe ople fou nd t hat the flui d ca n al so b e do ne. And then we star ted

995- 96. At t hat time pe ople fou nd t hat the flui d ca n al so b e do ne. And then we star ted

to t his day, we are stil l do ing vari ous simu lati on t opic s. T he i mpor tant goa l is to rest ore the phys ical eff ects of the real wor ld i n co mput ers. Of cour se, grap hics has a f eatu

re t hat it i s mo re i ncli ned to t he f ield of visu al e ffec ts. And then wit h th e em erge nce of t his tech nolo gy, some of them are als o us ed i n ro boti cs. Simu lati on i s do ing this

thi ng. Of c ours e t here are som e ot her bran ches of comp uter gra phic s s uch as c ompu tati

thi ng. Of c ours e t here are som e ot her bran ches of comp uter gra phic s s uch as c ompu tati onal pho togr aphy . Yo u pl ay a lot of came ras, you sho uld know tha t th ere are many ma ny s

tudi es i n th e wh ole imag ing post -pro cess ing. how to d o f or e xamp le, supe r re solu tion to inc reas e th e re solu tion of phot os. But ther e ar e ac tual ly a lot of. .. B ecau se i t us

es t echn olog y li ke t he s crol l-ba sed neur al n etwo rk, a lo t of peo ple will put it toge ther wit h CV and com pute r vi sion . So vis ion and grap hics hav e al ways bee n tw o se para

te s ubje cts. We know wha t gr aphi cs a re. I wa nt t o kn ow w hat Tai Chi is. Why do w e ne ed t o do Tai Chi ? Wh at a re t he i mpor tant pro blem s in Tai Chi to solv e gr aphi cs? The firs t ti

me, the main app lica tion of Tai Chi was simu lati on. I wa nted to use a se t of goo d ba sic faci liti es t o ma ke t he n eedl e be pus hed to a ver y la rge scal e. M y go al f or m y Ph D wa s to

run 11 part icle s of nee dle on a sin gle GPU. You did it? One bil lion par ticl es u sing Tai Chi . Th is i s co mple tely imp ossi ble befo re T ai C hi. Yes, com plet ely impo ssib le b efor e Ta

. Th is i s co mple tely imp ossi ble befo re T ai C hi. Yes, com plet ely impo ssib le b efor e Ta i Chi. Why not? Bec ause you wan t to ach ieve thi s th ing. .. Y ou w ant to u se a n ex ampl e of

i Chi. Why not? Bec ause you wan t to ach ieve thi s th ing. .. Y ou w ant to u se a n ex ampl e of 1 b illi on p arti cles . I reme mber it was a 30 90 G PU, and that GPU was 24G B of RAM . 24 GB o

f RA M me ans that the sto rage spa ce o f ea ch p arti cle can' t ex ceed 24 byte s. O nly by d oing thi s ca n yo u pu t 10 ^9 o n a piec e of RAM. In o rder to achi eve this I did a lo t of thi

ngs in t he e dito r. I rem embe r I coul dn't bea r to use the 16- digi t ne gati ve n umbe r. I

use d a part icle it s x, y, z. I wan ted to p ut x in a 32 -dig it i nteg er. So i t se ems that x u sed 11 d igit s y use d 11 dig its, and z o nly gave it 10 d igit s. S o 11 plu s 11 plu s 10 is

32. But if y ou d on't hav e a comp iler if you wri te t his your self it wil l be ver y v ery pain

32. But if y ou d on't hav e a comp iler if you wri te t his your self it wil l be ver y v ery pain ful. So I did a w ork call ed Q uant Tai Chi . I coul d us e al l th e da ta s truc ture s in Tai

ful. So I did a w ork call ed Q uant Tai Chi . I coul d us e al l th e da ta s truc ture s in Tai Chi to do q uant izat ion, to make a l ow b it d ata stru ctur e. T his way, a p iece of GPU coul

d fi t 11 par ticl es. So y ou c an s ee w hat Tai Chi is a bout . Ta i Ch i is abo ut i nfra stru ctur e. I t al lows our sim ulat ion to b e mo re e asil y de velo ped, run fas ter, sav e mo re m

ctur e. I t al lows our sim ulat ion to b e mo re e asil y de velo ped, run fas ter, sav e mo re m emory, and the envi ronm ent arou nd i t be come s mo re f rien dly and also dif fere ntia ble. So

it c an b e us ed i n ma ny r obot ics scen ario s. W hy d idn' t th e co mmun ity of g raph ics in t he p ast deca des make suc h ef fort s? Y es, actu ally a long tim e ag o I act uall y th ink

CUDA is such an effo rt. So C UDA is a ctua lly on C ++, it a lso uses the C++ gra mmar so tha t ev ery C or C++ pro gram mer can writ e GP U pr ogra ms a nd w rite par alle l pr ogra ms. So w e tr

ied to g et a s cl ose to C or C++ as p ossi ble. But as time wen t on we fou nd t hat many of the plac es w here C++ was bui lt w ere repl aced by high er-l evel lan guag es l ike Pyth on. Ther e we

re e ven many way s to wri te h igh- perf orma nce prog rams in Pyth on. The most typ ical cas e is PyT orch . In fac t P yTor ch i s a fram ewor k th at a buse s Py thon 's m any lang uage s a llow

PyT orch . In fac t P yTor ch i s a fram ewor k th at a buse s Py thon 's m any lang uage s a llow ing you to p erfo rm h igh- perf orma nce comp utin g in Pyt hon. The sam e go es f or T IG. Befo

re h igh- end lang uage s li ke P ytho n an d LV M we re m atur e b efor e co mpil er i nfra stru ctur e li ke L VM w ere matu re, befo re t he s imul atio n ap proa ch l ike MPM Befo re y ou b ecom

e ma ture be fore you bec ome matu re i n th e GP U pr ogra mmin g i t's hard to make a s plas h. Y

ou c an't cau se a n im pact . Yo u ca n't have a b ig i nflu ence . I thin k it was a g ood time . Fo r me I like to work on thes e th ings . Bu t yo u ma y ne ed a you ng m an w ho i s bo rn w ith

. Fo r me I like to work on thes e th ings . Bu t yo u ma y ne ed a you ng m an w ho i s bo rn w ith a go od s ense of humo r an d th inks he can make a g ame. The cou rage is also ver y im port ant.

So w hen you wrot e Ta i Ch i at MIT yo u di d it sol o. W hat was that lik e? I t wa s gr eat. I f

ound a p oint . I can do w hat I li ke, I ca n se nd a pap er, I ca n op en i t up I can let peop le u se i t I can mov e my self to a sm all room and not hav e to dea l wi th a nyon e. I can mak

e my men tor thin k " You did a go od j ob." um I wr ote a lo t of art icle s ab out this job in

clud ing PTX. I d id t he f ull stac k. I thi nk i t wa s a grea t ex erci se f or m e. L ooki ng b ack, I w as a pro duct man ager . I anal yzed the nee ds o f us ers. I w as a sim ulat ion engi neer

. I need ed t o kn ow w hat infr astr uctu re w as n eede d. I was a u ser, so I co uld beco me a

. I need ed t o kn ow w hat infr astr uctu re w as n eede d. I was a u ser, so I co uld beco me a pro gram man ager . I lear ned a lo t of ski lls that pro duct man ager s sh ould hav e. I als o pl

ayed the rol e of a s cien tist. to u nder stan d ho w th e la test tec hnol ogy can conv ert into a g ood prod uct for ever yone to use. I a lso play ed a goo d ro le o f an eng inee r i nclu ding

set ting up the GitH ub p roje ct, like CIC D t he e ntir e so ftwa re s truc ture an d th e pe rfor manc e op timi zati on o f ma ny C ++ l ayer s i nclu ding the mul ti-l ayer par alle l tr ansl atio

n t he t rans lati on c ache an d th e en tire des ign of t he m iddl e ta ble. The se a re a ctua lly the work of engi neer s. T hen I di d so meth ing that a C EO s houl d do wh ich is t o ma ke t

hese thr ee r oles 做的这个 成果整合 起来 让 更多人的 用上 所 以相当于 就是某一 种叠加态 或者是 一个雌雄 同体的一 种状态 也好就自 己闭环 把这个事 情做出来 了 我能

想象这是 一个 听 起来非常 浪漫的 个人英雄 主义的P hD的故 事 对 特别是在 波士顿 因为可能 下三四个 月 还是 在COV ID期间 对吧 四五个月 下雪 你 在自己房 间里面 外面飘着

鹅毛大雪 然后你 在里面 既当工程 师 又当 签名请 又当科学家 Look ing back at the past few yea rs, are ther e an y mo ment s th at t ouch ed y ou? Or a re t here any mom ents tha t yo u th ink Tai Chi can chan ge t he w orld

? Fi rst of a ll, I do n't thin k I need tha t ki nd o f ti me t o dr ive myse lf t o co ntin ue d

? Fi rst of a ll, I do n't thin k I need tha t ki nd o f ti me t o dr ive myse lf t o co ntin ue d oing thi s. O f co urse th ere are a fe w mo ment s th at I fin d pa rtic ular ly i nter esti ng.

For exam ple, whe n I firs t us ed m y ow n pr ogra mmin g la ngua ge t o im plem ent my o wn p rogr am, and then I f ound it was bett er t han CUDA . I was so s hock ed. You see, Hua ng f ound so many

peo ple to c reat e su ch a pro gram ming lan guag e b ut t here was not eve n on e PH P st uden t to mak e it . Bu t it was onl y us eful in a ce rtai n fi eld. I t houg ht i t wa s ve ry i nter

esti ng. Anot her mome nt w as w hen I ad ded Auto Diff to the prog ram. I s pent abo ut 3 -4 d ays

esti ng. Anot her mome nt w as w hen I ad ded Auto Diff to the prog ram. I s pent abo ut 3 -4 d ays to t urn the prog ram from "fo rwar d" t o "b ackw ard" . I was in A dobe 's t ime, not far fro m he

re. I re aliz ed t he f irst MTM -def inis habl e so ft b ody simu lator. uh This is a ve ry i nter

re. I re aliz ed t he f irst MTM -def inis habl e so ft b ody simu lator. uh This is a ve ry i nter

esti ng t ime. I a lso thin k th at o nly if y ou h ave a si mula tor can you achi eve this . Yo

u ju st m enti oned a k ey w ord, bec ause Tai Chi is actu ally an open sou rce proj ect. I a m ve ry c urio us a bout you r op inio n on ope n so urce . Be caus e Ta i Ch i is a v ery good ope n so

urce ph ysic al s imul atio n la ngua ge. And now ther e ar e op en s ourc e IO Ms l ike LAMA . Th

ey a ll s eem to h ave a co nsen sus, Righ t. I 'm c urio us a bout you r op inio n on ope n so urce . Wh at s houl d a star tup comp any do t o op en s ourc e? I 've had this opi nion sev eral tim es.

. Wh at s houl d a star tup comp any do t o op en s ourc e? I 've had this opi nion sev eral tim es.

I've bee n th inki ng a bout whe ther my judg ment on this iss ue i s co rrec t. O f co urse I was good at open ing sour ce i n th e ea rly days . Wh en I was at MIT, man y gr aphi cs p aper s we

re n ot o pen sour ce. Ther e we re n o co des adde d to the m. I did one thi ng. All the resu lts of a ll m y pa pers cou ld b e sh own in o ne o rder. If y ou l ook at t he c ode of t he T aiji yo

u wi ll f ind that all of its expe rime nts have a c omma nd l ine. The n yo u ru n th is c omma nd l ine, and you can sho w th e re sult to it. This is when I w as d oing my PhD. I t houg ht t hat

open ing a co mpan y wa s a grea t th ing. Aft er g radu atin g fr om t he P hD, I st arte d wo rkin g in a c ompa ny. At f irst I thou ght that mat ure open ing proj ects We c an a lso do a lot of

thin gs l ike Data bric ks. They sta rted as a re sear ch p roje ct a nd n ow t hey are one of t he l arge st r esea rch proj ects wit hout a l isti ng. When I s tart ed w orki ng i n th e co mpan

y I wan ted to c onti nue work ing in T ai C hi. I al so d id a lot of arti cles and did a l ot o f co mmun ity work . Th ere are more and mor e pe ople usi ng i t. B ut l ater I enco unte red some

cha llen ges in c omme rcia liza tion. At t hat time I thou ght, OK, is it a mat ter of l ife and deat h fo r a busi ness com pany to open a c ompa ny? Beca use you gave you r be st t o ot hers ho

w ca n yo u ma ke m oney ? Th is i s a very rea l pr oble m. S o i n fa ct, for a wh ile, Tai ji's res earc h an d de velo pmen t in vest ment was rel ativ ely smal l. A nd t hen, in fact we als o ha

d a lot of e xper ienc e in doi ng t he n ew p roje ct o f ME SHI. So, Tai ji's Li Bo t rans ferr ed t o ME SHI. And then as time wen t on to tod ay, beca use toda y i n fa ct, Manc huri an b usin

ess has done qui te w ell. We have a g ood reve nue. I a lso have som e ex tra ener gy t o th ink abou t th e is sue of r esch edul ing. It is t rue that res ched ulin g do es n ot m ake mone y. B

ut y ou h ave a go od r esch edul ing proj ect that can hel p yo u at trac t th e be st t alen ts a nd t he b est cust omer s. I t's equi vale nt t o it bec omin g a mark etin g me thod. Peop le w

ho a re t echn ical may dis agre e wi th t he t erm "mar keti ng" but in m any indu stri es, rese arch and mar keti ng a re e qual ly i mpor tant . Fo r ex ampl e i n th e ga ming ind ustr y i f I

spen d 10 0 mi llio n RM B on a g ame, it' s no rmal for me to s pend 100 mil lion RMB on mark eting.

That 's g reat. And as a tec h st artu p I 'm n ot g ood at t radi tion al m arke ting . I' d ra

ther do a go od p rodu ct, do t his kind of tech nica l ma rket ing, and bui ld t rust by givi ng y ou t he c ode to l et y ou k now me. So a ctua lly, Kai yuan has its pla ce i n a star tup comp any.

但是这个 得非常s trat egic 。 对, 得非常s trat egic 。 你得 想到一个 好的商业 化的方式 和它互补 这样 你整个事 情才能够 闭环地做 下去。 要不然很 快大家就

会问你, OK, 我们为 什么要有 一个部门 在做开源 ? 也会 影响你整 个公司的 竞争力, 但是你 用得好的 话 它 是能增强 整个公司 的竞争力 的。 但 是我觉得 最根本的 事情就是 对我

来说我还 是很热爱 开源的, Many tim es, why a pe rson sta rts a bu sine ss i s be caus e he wan ts t o do som ethi ng t hat is i deal for a g roup of peop le. Sinc e th e so urce of the sour ce i s in my idea

l i t wi ll d efin itel y be in vari ous form s. Y ou c an't thr ow i t aw ay u ntil you may wal k to the sou rce again. Ther e ar e tw o la bels in Tai Chi, one is "Kai yuan " an d th e ot her

is " Phys ical Sim ulat ion" I'm cur ious if we want to make phy sica l si mula tion mor e us eful for rob otic s w hat else do we n eed to d o? W hich dir ecti on i s th e hi ghes t pr iori ty? That

's a goo d qu esti on. I th ink that in some sen se, the curr ent simu lato r ne eds to b ecom e mo re d ata- driv en. Neur al, the way of t he n eura l ne twor k is com bined. This doe sn't mea n th

at t he N ewto n's law of N ewto n's law is i nacc urat e. T he N ewto n's law of N ewto n's law is a ccur ate. Rel ativ e th eory is also acc urat e. B ut m any boun dary con diti ons may not be a

ccur ate. Man y of the hoo ks, fric tion law s i ts m odel may not be accu rate . Be caus e th ere

ccur ate. Man y of the hoo ks, fric tion law s i ts m odel may not be accu rate . Be caus e th ere are too many dif fere nt f rict ion in t he w orld . Th e fr icti on m echa nism of the ligh t su

rfac e an d th e li ght surf ace and the fric tion mec hani sm o f pu ttin g a glas s on a w et s urfa ce a re c ompl etel y di ffer ent. The se m odel ing are all inac cura te. So t hese gap s be

twee n si mula tion and rea lity nee d to be clos ed i n a data -dri ven way. On the othe r ha nd, alth ough eve ryon e is now foc usin g on end -to- end, you can sti ll s ee t hat ther e ar e ma

ny i nduc tive bia ses adde d to the se e nd-t o-en d ne ural net work s. T he m ost clas sic exam ple is t he r ope in t he T rans form er t hat has been use d mo re r ecen tly, Rot ary Posi tion

al E mbed ding. It's a r elat ive posi tion al e mbed ding tha t is rea lize d th roug h th e mu ltip lica tion . If I l eave out the se i nduc tive bia ses, I d on't eve n ne ed a tra nsfo rmer I'

m ju st a n ML P f rom begi nnin g to end it 's a ll d ense con nect ion. Is that oka y? I f yo u ha ve e noug h co mput e e noug h da ta, it's pro babl y ok ay. But in r eali ty, you don' t. S o yo

u ca n on ly i njec t in duct ive bias fro m so me p hysi cal thre e-di mens iona l wo rlds int o yo ur l earn ing syst em. I th ink this is a pa rtic ular ly g ood plac e to com bine phy sica l si

mula tion and neu ral netw orks or AI. Simu lati on r equi res more AI. AI also req uire s mo re 3 D in duct ive bias. We n eed to f ind a ba lanc e po int that is just rig ht, not to b e to o ha

rd o n th e le arni ng p roce ss, not to u nder esti mate the lea rnin g ab ilit y of the mod el.

But at t he s ame time we don 't r eall y gi ve e very thin g to lea rnin g. A lot of the unde rlyi ng i nsig hts are very hel pful for the sys tem to l earn. If w e wa nt t o tu rn t he c urre nt-s

tage pro babi lity and the sec ond- orde r pr obab ilit y of the rul e-ba sed simu lato r in to l earn able wh ich part do you thin k sh ould be trai ned? Wha t ki nd o f da ta? Let' s ta lk a bout

a u se s cena rio firs t. I thi nk t his migh t be a b ette r ex ampl e. F or e xamp le, if y ou a re d oing vid eo g ener atio n y ou m ight nee d a came ra m odel . If you can do a vi deo simu

lation Thes e ob ject s ar e no t ex pres sed usin g pi xels bu t ra ther usi ng a 3D repr esen tati on. Then you har dcod e th e ca mera pro ject ion. Can you sav e a lot of p aram eter s to lea rn c

on. Then you har dcod e th e ca mera pro ject ion. Can you sav e a lot of p aram eter s to lea rn c amer a pr ojec tion ? In fac t c amer a pr ojec tion is some thin g th at c an b e do ne i n tr adit

iona l gr aphi cs s uch as 5 0 fl ops. But if you use AI t o do it, you may hav e to add a b illi on p aram eter s. S o in duct ive bias is very ve ry e ffec tive in this. Or n ow, when we do v ideo

mod elin g w e of ten enco unte r a prob lem, tha t is if I t urn the came ra 3 60 d egre es, I se e a comp lete ly d iffe rent wor ld. Do y ou h ave some bas ic 3 D mo deli ng w ith Part icle or

Voxe l in thi s wo rld? At leas t yo u ca n so lve the phys ical per sist ency pro blem. If y ou l ook at t he s imul ator s we can use on the mark et, from Isa acs, Moj oko, Dra ke, PyBu llet Ge

nesi s a ll t hese sim ulat ors, wha t ar e th eir prob lems ? I thin k th e co mmon pro blem is the sim to r eal gap. Bec ause the se s imul atio ns, of c ours e h ave cert ain pred icti ons. But in

fact I thin k ev eryo ne w ho u ses simu lato rs t o do rob otic s sh ould kno w th at y ou h ave to m atch the se p aram eter s y ou h ave to m atch the se b ound ary cond itio ns, and then let the

se s olve rs p lay 最好的状 态其实是 相当难的 一件事情 。 可能 在未来的 simu lato r里面, 它当然 会有刘顿 第二定律 没有 意义说我 用一个t rans form

er 去 重新把刘 顿第二定 律选一遍 我用 etfl ops去 pred ict一 个它下一 帧的位置 本来 可以ef lops 就做完, 做好的一 个事情, 你用e tflo ps去做 ,那当然

不mak e se nse, 但是你 也考虑到 整个物理 世界的复 杂性, you can' t us e th e Ne wton 's l aw o r th e Eu ler' s la

w or Nav i-St okes to mode l. T hey are all simi lar. At this poi nt, you may need a l ot o f le

arna ble syst ems to h elp you capt ure the work of thes e va cuum -env iron ment phy sica l la ws.

In r eali ty, ther e is air res ista nce, the sur face of the obje ct i s no t a steel.

Or t he b ound ary cond itio ns a re n ot e xact ly t he s ame as y ou t hink . Yo u ca n't meas ure the boun dary con diti ons. How can suc h a syst em h elp your tra inin g as muc h as pos sibl e? I

thi nk t his is a ver y in tere stin g ph ysic al s imul ator and neu ral deep lea rning. 要结合的 地方。其

实太极如 果朝这个 方向发展 的话,有 很多事情 可以去做 。 你如 果要做这 个事儿, 你就必须 要有一个 unde rlyi ng的i nfra stru ctur e 使 得它可以 非常高性

能地继续 像MPM 一样去解 它的el asto plas tici ty。 与此同时 ,它的整 个材料可 能能够, 比如说它 的str ess stra in c urve 这样 的str ess

stra in c urve 是不是可 以做到d ata driv en, 那你就需 要高性能地 We a re t alki ng a bout the dat a dr iven . I am c urio us a bout the lim it o f ph ysic al s imul atio n. H ow f ar c an w e si mula te t he w orld

? If we can perf ectl y si mula te t he w orld ca n we cre ate AGI? Whe re i s th e li mit? Whe n do

? If we can perf ectl y si mula te t he w orld ca n we cre ate AGI? Whe re i s th e li mit? Whe n do we know it is w orth doi ng? I th ink it i s a good que stion. 一个很伤 心的事实 是,当然 这是我的 一个猜想 ,如果说 我们自己

的这个世 界就是模 拟出来的 话,一个 模拟器里 面的另外 一个模拟 器,它是 不会比它 的母体模 拟器更强 的。 所 以我们永 远也不会 。 所以 我觉得也 许你在我 们的世界 里面去造 一个模拟

器,模拟 整个世界 ,使得这 个模拟器 运行速度 能比现实 的时间更 快,这个 有可能是 unli kabl e,除非 你去模拟, just do a sm all scal e si mula tion . Th is a lso mean s th at i t's prob ably har d to pre dict the sto ck m arke t. B ecau se t

tion . Th is a lso mean s th at i t's prob ably har d to pre dict the sto ck m arke t. B ecau se t he s tock mar ket invo lves all the peo ple in t he w orld inv olve d in thi s th ing. But you can

do a wi nd-u p si mula tion . Wh at i s it s li mit? To be h ones t t he c urre nt s imul atio n is dev elop ing. .. I sho uld say it's fac ing a bo ttle neck . In the pas t p eopl e mi ght have fou

nd t hat the fina l el emen t t he f inal vol ume, or the mate rial poi nt m ethod or t he m ass- brin g sy stem it alw ays has some new dis cret izat ion sche mes or s ome acce lera tion met hods

tha t ca n be con stan tly prop osed . In fac t i n re cent yea rs, the prog ress in this are a ha s be en r elat ivel y no t as muc h as bef ore. So I ac tual ly t hink tha t th e pu re v alue of

this nee dle, of cour se, it h as a lon g us e hi stor y l ike Simo ns o r like Flue nt, such a s oftw are, is actu ally use d in man y sc enar ios. I a ctua lly feel tha t th is t

echn olog y is a b it o n th e sa me l evel . If we go f urth er, the most pro misi ng d irec tion is to p ut s ome AI s tuff int o it . Pu t th e... For exa mple,

elas topl asti city con trol is dead you wil l fe el t hat simu lati on i s al so k ind of d ead

Ther e is a l ot o f pr ogre ss i n th e se gmen ting fie ld. But in t he w hole fun dame ntal th e bi g br eakt hrou gh, at l east I d idn' t he ar t oo m uch. I s aw y ou w rite tha t wh en y ou g

radu ated fro m MI T y ou t houg ht t hat grap hics was als o in the bot tlen eck. You tho ught you did n't know wha t to do in t he f utur e. L ooki ng b ack, do you thin k yo ur j udgm ent was corr

ect? I t hink so. At that tim e w hen ever yone was doi ng r ende ring it was alr eady ... Actu ally

ect? I t hink so. At that tim e w hen ever yone was doi ng r ende ring it was alr eady ... Actu ally

I thin k th at 2 0 ye ars ago, man y th ings abo ut r ende ring had alr eady bec ome some sor t of tec hnic al a rran geme nt. For exam ple, I r emem ber ther e wa s a algo rith m ca lled AMC -MC-

PPM, whi ch i s se lf-a dapt ed t o th e Ma rkov cha in, Mont e Ca rlo, and uni tes the path sam ple or t he t op p oint to the merg er. If y ou t rans late it into Chi nese. It c ombi nes MCMC Un ifie

d Pa th S ampl ing, and Ada ptiv ity. It' s a good and int eres ting job . Bu t wi th t his kind of job, it mean s th at t here won 't b e mu ch p rogr ess. I t hink thi s is a v ery inte rest ing

rule . Wh en y ou f ind a pr oble m th at i s ge ttin g cl eane r an d mo re w ell- defi ned, it

rule . Wh en y ou f ind a pr oble m th at i s ge ttin g cl eane r an d mo re w ell- defi ned, it mean s th at t here 's n othi ng y ou c an d o ab out it. When it form s a spec ial syst em b ench

mark, When it beco mes more and mor e co mpli cate d i t me ans that the re i s le ss c hanc e to get thr ough thi s. Y ou j ust talk ed a bout all the sim ulat ors, and you thi nk t hey all have

sim 2rea l pr oble ms. Do y ou b elie ve i n si m2re al? I be liev e. I bel ieve . I thin k al thou gh t here are som e pr oble ms, simu lati on s till has a g reat mea ning . Do n't say that the

whole - Righ t? B ecau se i n th e ai r t o be hon est, the sky is just air . At mos t i t's a bi

t cl oudy or some thin g. B ut w hen you want to mani pula te t hing s mo re c aref ully it 's a ctua lly very ve ry d iffi cult . An d th ere are many unb alan ced boun dary con diti ons in t he m

iddl e w hich wil l le ad t o so me s imul atio ns i n re alit y. I t's very ve ry d iffi cult to

land . Th ere must be vari ous adap tati ons. And it' s no t li ke... -

land . Th ere must be vari ous adap tati ons. And it' s no t li ke... -

Serg ey L evin e fr om B erke ley wrot e a blog cal led "The Spo rk o f AG I" T he m ain poin t is tha t he thi nks that usi ng s imul atio n is a l imit atio n fo r hu man bein gs t o us e th e Ro

boti c Fo unda tion Mod el. He e xpre ssed his opp osit ion to s imul atio n in tha t bl og. He t

hink s th at w e sh ould use rea l-wo rld data to crea te t he n ext gene rati on o f fo unda tion rob otic mod el. We s houl d no t us e sy nthe tic data or simu lati on. How woul d yo u ev alua

te t his? I t hink it make s se nse. Bec ause syn thet ic d ata can only be used to enha nce a mo del. It can' t bu ild a mo del from scr atch fr om t he b otto m up . Be caus e if you wan t to tra

del. It can' t bu ild a mo del from scr atch fr om t he b otto m up . Be caus e if you wan t to tra in a fou ndat ion mode l u sing syn thet ic d ata, all dat a is syn thet ic d ata, or most dat a is

syn thet ic d ata, it will not wor k. B ecau se t hese so- call ed s ynth etic dat a me ans you can synt hesi ze it. Once you can syn thes ize the data yo u mu st f ollo w ce rtai n ru les. The se r

ules mus t be tho ught out by huma n be ings . Le t's not talk abo ut t he c over age of t hese rul es. In t he r eal worl d y ou c an o nly cove r 10 ,000 rul es. Let' s no t ta lk a bout the act ual

es. In t he r eal worl d y ou c an o nly cove r 10 ,000 rul es. Let' s no t ta lk a bout the act ual phys ical law. Give n th is a rgum ent, do you thin k a gene ral robo tic simu lati on w ill be e

stab lish ed? Can we c reat e a robo tic simu lati on t o he lp u s do thi s ki nd o f th ing? I t

hink if you comb ine simu lati on w ith VLM and othe r fo unda tion mod els, it will be esta blis hed. Do you thin k th ere is a cha nce? I t hink the re i s a chan ce. Of c ours e w hat you get

hed. Do you thin k th ere is a cha nce? I t hink the re i s a chan ce. Of c ours e w hat you get is n ot a sim ulat or o r a neur al n etwo rk. What is it? It i s a 3D i nduc tive bia s th at i

s in volv ed i n th e ne ural net work . It is an i nduc tive bia s th at m akes you r ne ural net work eas ier to l earn . Bu t it is also dif fere nt f rom your own neu ral netw ork. It m ay c onta

in a var iety of magi cal oper ator s. F or e xamp le, I ca n pi ck u p a numb er o f po ints in a 3D spa ce. It c an b e an ti-e nd d iffe rent iabl e. I thi nk i f it 's l ike this yo u ca n ac

tual ly p ut t he b est simu lato r in the sim ulat or, like I t hrow a b all from spa ce, its traj ecto ry i s a thro w li ne, data dri ven meth od f or t his kind of stro ng a dapt ivit y un der

comp lex bord er c ondi tion s. Y ou j ust ment ione d a lot of l imit atio ns, like the lim itat ions of the curr ent simu lato r. I f Hu Yan ming wan ts t o re make a s imul ator tod ay, what wil

l he do? Wha t wi ll h e do ? I thin k ma ybe I wi ll u se l earn ing to d rive it, rat her than usi ng t he f low rate law to driv e it. Beca use I di d a lot of t hese ba sed on N avie r-St okes ba

sed on E uler -Hal ag-L ang' s eq uati on, base d on New ton' s la w of law or aft er d oing the se s imul ator s I fou nd t hat thes e ph ysic al l aws are all appr oxim atio ns. Newt on's law may

not be, but man y ot hers th ey t hems elve s ma y no t be app roxi mati ons, not a m irro r. B ut t he e nvir onme nt i t de scri bes, its bou ndar y co ndit ion, are all mir rors. So f rom this per

spec tive I woul d on ly w ant to k eep some of the more und erly ing elem ents and mak e al l of the se p hysi cal laws int o da ta d rive n. I nclu ding all of the defo rmab le e leme nts? Yes th

e mo re u nder lyin g el emen ts i nclu de t hat I ha ve a cup on the tabl e i t wo n't disa ppea r in the nex t se cond . In clud ing two obje cts that can 't o ccup y th e sa me p osit ion. The

se m ust be k ept? The se m ust be k ept. Inc ludi ng I mig ht h ave a ca mera . Th e ca mera 's l ens is a pers pect ive proj ecti on. So I can lea rn h ow t o op erat e th ese obje cts in t he r

eal worl d. H ow t o ch ange the m. S o I can lear n fr om v ideo s. A nd I don 't n eed to u se c ompl ex s enso rs t o ca ptur e ea ch o bjec t's posi tion it s ro tati on, its traf fic, etc . Do

you sti ll n eed to k eep rend er a nd r ay t raci ng? I th ink afte r a long tim e r ay t raci ng w ill be r epla ced by n eura l gr aphi cs.

Actu ally yo u ca n al read y se e th is t rend ri ght? Now wi th t he D LSS tech nolo gy, a 4K scr een can only gen erat e 1K it onl y ge nera tes 1/8 or 1 /32 pixe ls. And each pix el's lig ht m

ay o nly hit one samp le. Then you can use the nee dle or t he s urro undi ng s ampl e to gen erat e it in a da ta-d rive n way. Afte r a long tim e y ou c an a ssum e th at I exp ress the who le s

peed of the spac e fo r a neur al n etwo rk. It c an a ctua lly rend er t his thin g in a 3 D Tr ansf orme r wa y wi thou t re trac ing. Why do we s till use TIG to rend er a nd d o ph ysic al s

imul atio n? W hy d on't we use Sora ? Or sho uld I sa y c an w e so lve vide o ge nera tion and 3D plan ning ? It 's d efin itel y po ssib le. Beca use, as I sa id, TIG is i n th e fu ture . Of cou

plan ning ? It 's d efin itel y po ssib le. Beca use, as I sa id, TIG is i n th e fu ture . Of cou

rse, we are plan ning how to do T IG 2 .0. It's an infr astr uctu re s uppo rtin g gr aphi cs a nd A I. B oth side s ne ed t o su ppor t it . Do you rea lly need a m ore and more pre cise sim ulat ion?

I. B oth side s ne ed t o su ppor t it . Do you rea lly need a m ore and more pre cise sim ulat ion?

For exa mple 3D lim iter s w ith high er a nd h ighe r nu mber of node s a nd m ore and more com plex sco re s chem es, and more and mor e di ffic ult to s olve . I don' t th ink this is the best

dir ecti on. The best dir ecti on s houl d be I' ll j ust acce pt i t. T hese bou ndar y co ndit ions th ese phys ical laws, Or t hese thi ngs that Gal erki n ma de u p l ike weak for m a re n ot t

he b est way to d escr ibe the phys ical wor ld. Beca use the phys ical wor ld i s to o co mple x. Y

ou h ave to u se d ata- driv en m etho ds t o de scri be t hese par ticl es, how they mov e. S o ac tual ly, five yea rs a go, your 99- line wri ting was ice -col d. N ow, you can only wri te 2 0 wo

rds for all the prom pts. Do you thin k th is n ew v ideo mod el i s th e fa te o f th e ol d gr aphi cs p eopl e? S ome scen es a re, yes. But I th ink a lo t of thi ngs that gra phic s do es a

ctua lly prom ote the birt h of Sor a. W ith the fina l ev olut ion, I t hink lea rnin g wi ll b e th e ma in t hing in the futu re. But the simu lato r wi ll, in s ome form th roug h in duct ive

bias th roug h sy nthe tic data it wil l be come a p art of t he l earn able mod ule. We just tal ked abou t a few conc epts. One is c alle d th e re al w orld . Th e ot her is t he w orld we can

obse rve with our eye s. T he t hird is the worl d we can mod el w ith vari ous kind s of phy sica l mo dels . Do you thi nk t he g ap b etwe en t he t hree is big? Ver y in tere stin g. I don 't k

now what the rea l wo rld is l ike. The wor ld w e ca n ob serv e ac tual ly g oes back to the prob lem we t alke d ab out earl ier. Whe n I was slee ping th e tr ee d owns tair s w as i t re ally gro

wing ? Or was it just a m omen t wh en I cou ld o bser ve i t a nd i t in stan tly grew int o wh

wing ? Or was it just a m omen t wh en I cou ld o bser ve i t a nd i t in stan tly grew int o wh at i t sh ould be? But we c an o nly obse rve what we can obse rve. Thi s is a v ery smal l pa rt o

f th e wo rld. So ever yone 's p erce ptio n ha s a huge lim it. I th ink the two, the rea l wo rld and the worl d we can obs erve ha ve a ver y v ery big gap. The wor ld w e ca n ob serv e an d th

e wo rld that phy sica l si mula tion can det ect are anot her 180, 000 mile s aw ay. But the gap is d iffe rent fro m th e pr evio us o ne. The prev ious one was tha t ev eryo ne c ould onl y ob

serv e wh at t hey coul d ob serv e. T he s econ d ga p wa s th at w e di dn't hav e en ough com puti ng. We d idn' t ha ve e noug h da ta t o mo del what the rea l wo rld was like . So me c lass

puti ng. We d idn' t ha ve e noug h da ta t o mo del what the rea l wo rld was like . So me c lass ic a ppli cati ons are actu ally pos sibl e. F or e xamp le, I bu ilt an a irpl ane, and I b lew

a wh istle. usin g CF D to cal cula te t he r esul t i t mu st b e ve ry s tron g. W heth er y ou c an t hrow the win d tu nnel or not, you mus t bl ow t he w ind tunn el b efor e ta king off . Yo

u ca n't say that CFD is over . So I t hink it' s he lple ss. As a spe cies ma ybe oper atin g in ano ther sim ulat or, we e ven drea m of cre atin g ou r ow n si mula tor, whi ch c an s imul ate

ours elves. I th ink it's a m irac le t o se e so me G PT-4 mod els that can mim ic t he o pera tion

ours elves. I th ink it's a m irac le t o se e so me G PT-4 mod els that can mim ic t he o pera tion of huma n br ains . Sp eaki ng o f GP T-4, I w ant to a sk y ou, do y ou t hink tha t a mode l li

ke G PT-L OM l ives in the cave of the Bola Tou mode l? Y es, beca use its loss fun ctio n is its

tra inin g da ta. Ever ythi ng i t do es i s to min imiz e th is l oss func tion. 嗯,特别 好的问题 。 我觉

得...

It m ay n ot r equi re a cer tain flo w ra te a nd L agra ngia n fu ncti on, but it m ay r equi re t ool use. For exa mple it may req uire its own com pute r. F or e xamp le, AGI can writ e a code lik

ool use. For exa mple it may req uire its own com pute r. F or e xamp le, AGI can writ e a code lik e "f ine elem ent" to help ana lyze wha t it loo ks l ike when a p iece of rubb er i s dr oppe d on

the gro und. Why a p iece of Ital ian doug h is dro pped on the tabl e an d it wil l be bro ken into thr ee p iece s? I t ca n so lve thes e pr oble ms w ith "fin e el emen t" o r ot her meth ods usin

g to ol u se. Why can' t it sol ve i t by its own nex t to ken pred icti on m etho d? B ecau se t he n ext toke n pr edic tion met hod has a li mit in e ssen ce. No m atte r ho w li ght, hea vy,

or l arge the tok en i s i t mu st s pend twi ce t he p aram eter cou nt/s parc ity/ MOE. The AGI

I'm tryi ng t o do now is to m ake the scal e bi gger and big ger. Thi s wi ll m ake the numb er o f pe ople who can com pute and evo lve each tok en m ore and more . Bu t ma ny t oken s ca n't pass

thr ough . Ev en i f yo u ha ve a lot of comp uter s s ome toke ns n eed more com puti ng t han othe

thr ough . Ev en i f yo u ha ve a lot of comp uter s s ome toke ns n eed more com puti ng t han othe rs. Thes e co mput ers can' t be in the LLM mode . Th ey n eed to b e ma nipu late d by LLM . So the

rs. Thes e co mput ers can' t be in the LLM mode . Th ey n eed to b e ma nipu late d by LLM . So the

too ls y ou'r e ta lkin g ab out are actu ally phy sica l to ols. May be t hey' re c ompu ting too ls. They nee d to hav e th eir own comp uter . Th en t hey can use thei r ow n Py thon in the comm

ls. They nee d to hav e th eir own comp uter . Th en t hey can use thei r ow n Py thon in the comm and line . Th en t hey can writ e th eir own thin gs i n th e co mman d li ne. And do y ou t hink

and line . Th en t hey can writ e th eir own thin gs i n th e co mman d li ne. And do y ou t hink thi s pr oces s do esn' t ne cess aril y re quir e so muc h re cogn itio n fr om h uman s ab out the

phys ical wor ld? It c an d isco ver its own prob abil ity. And can suc h AG I be bor n fr om t he t ext data vi deo data an d im age data we have now ? I thin k it can . Ok ay, my n ext chap ter is M

ESHI . Bu t be fore we talk abo ut M ESHI th ere' s on e la st q uest ion. Wil l yo u wa nt t o do

ESHI . Bu t be fore we talk abo ut M ESHI th ere' s on e la st q uest ion. Wil l yo u wa nt t o do rob otic s? I thi nk I mig ht. But I do n't know wha t I can do f or r obot ics. So f ar. I th ink

you can cont ribu te a lot . I thin k ro boti cs i s a very big imp act. May be t he o nly reas on I did n't star t do ing robo tics is beca use I'm not that int eres ted in i t to day. And I b elie

ve I can onl y do wha t I like. If o ne d ay I fin d Ro boti cs i nter esti ng, then we migh t go for it. Wha t ki nd o f th ings wil l ma ke y ou f ind Robo tics int eres ting ? Wi ll y ou c hang

for it. Wha t ki nd o f th ings wil l ma ke y ou f ind Robo tics int eres ting ? Wi ll y ou c hang e yo ur i nter est in i t? I don 't k now yet, so I'm not part icul arly int eres ted. So the resp

onsi bili ty i s on us. Yes we 'll try to g et m e in to t he R obot ics pit as s oon as p ossi ble.

Yuan ming no w th e bu sine ss o f Me sh i s ve ry s ucce ssfu l. W ith a $5 0 mi llio n ra ise and a $2 .5 m illi on t urno ver, you hav e cr eate d mo re t han 20 m illi on 3 D mo dels . I' m cu

a $2 .5 m illi on t urno ver, you hav e cr eate d mo re t han 20 m illi on 3 D mo dels . I' m cu riou s y ou s tart ed w riti ng T ai C hi i n th e do rmit ory at M IT. When did you rea lize you

wan ted to s tart a s tart up? Did you ever hes itat e or wha t wa s yo ur f eeli ng w hen you deci ded to s tart a s tart up? That 's a ver y go od q uest ion. I s till hav e a char acte rist ic.

I st ill like to try new thin gs. Thes e ne w th ings inc lude doi ng g raph ics and then doi ng t he e dito r. T hen doin g th e ed itor and the n do ing the diff eren tiab le p rogr ammi ng m odel

. Th at's how Dee pTou ch c ame about. When I w as d oing my PhD, I a ctua lly read a l ot o f in

. Th at's how Dee pTou ch c ame about. When I w as d oing my PhD, I a ctua lly read a l ot o f in terv iews of Chin ese and fore ign entr epre neur s. F or e xamp le, Elon Mus k L ei J un, Wang Chu

anzh i e tc. I th ink they are all ver y in tere stin g. W hen I wa s ab out to g radu ate, the re w ere a fe w op tion s. E ithe r I went to scho ol a s a teac her, or I ju st w ent to a com pany. The

main pro blem was tha t th e fi eld of s tudy was not as open as it w as. I di dn't kno w th at t here was suc h a good thi ng a s JN I. S o ba sica lly, I d idn' t kn ow w hat to d o wh en I gra

duat ed. So I tho ught OK ca n I make the thi ngs that the doc tor is d oing int o a comm erci aliz

duat ed. So I tho ught OK ca n I make the thi ngs that the doc tor is d oing int o a comm erci aliz ed p rodu ct? Then the re w as t he f irst Tai ji T u Xi ng c ompa ny. Late r w e ch ange d ou r na

me t o Ma sh b ecau se t here wer e mo re c usto mers ove rseas. So y ou d idn' t co nsid er a noth er w ay a t al l w hich was to work in a bi g fa ctor y. I cou ldn' t st and it. Why? I k new it w

ould be easi er t o wo rk i n a big fact ory. But I c ould n't stan d it . I had a bo ss. You had

to b e a boss . I didn 't m ind bein g a boss . I coul dn't sta nd i t. S omeo ne t old me w hat to do. But you also had a m ento r du ring you r Ph D. B ut h e wa s mo re l aid back . Ye s m y Ph

to do. But you also had a m ento r du ring you r Ph D. B ut h e wa s mo re l aid back . Ye s m y Ph D me ntor . Ev ery week I woul d or gani ze a ll t he t hing s I need ed h is h elp with a d ocum

ent. Dur ing the meet ing, I w ould go thro ugh the docu ment wit h hi m. A nd t hen mayb e we wou

ent. Dur ing the meet ing, I w ould go thro ugh the docu ment wit h hi m. A nd t hen mayb e we wou ld s pend a s hort tim e to geth er e very wee k. M aybe hal f an hou r. A nd t hen when we enco

unte red COVI D t he t ime woul d be sho rter . Bu t th e he lp h e ga ve m e wa s ve ry, very big . Wh ethe r it was enc oura ging me when I w as i n tr ouble, What kin d of com pany is Mesh 's b

. Wh ethe r it was enc oura ging me when I w as i n tr ouble, What kin d of com pany is Mesh 's b

igge st c lien t? B efor e I met Henr y I alw ays thou ght - Th e bi gges t cl ient sho uld be a gam ing comp any. - N o t he b igge st c lien t is ... Beca use if y ou l ook at t he p rice of the card

th e bi g co mpan ies may have som e AP I re quir emen ts. This is actu ally a l ot b igge r th an g amin g co mpan ies. You 've had a lo t of con tact wit h ga ming com pani es, righ t? G amin g st

udio s. T hey real ly n eed your 3D asse ts. I sa w yo u wr ite that the cos t of 3A game s is now

udio s. T hey real ly n eed your 3D asse ts. I sa w yo u wr ite that the cos t of 3A game s is now 50% on 3D a sset s. Y es. You' ve d ealt wit h so man y co mpan ies. Wha t do you thi nk i s th e bi

gges t pr oble m fo r th e ga ming ind ustr y? T he g amin g in dust ry h as b een shri nkin g in the pas t tw o ye ars. Alt houg h th ey a re a mar ket with a m arke t va lue of n earl y $3 00 b

illi on, you' ve s een a lo t of tal ent in t his indu stry . Ev eryo ne i s bu sy w ith the pric

e in crea se, 把以前老 游戏做个 重制版再 放出来。 其实整 个游戏行 业现在是 比较st rugg le的。 为什么 ? 有很 多原因, 就包括像 TikT ok这样 的短视频

的soc ial medi a的平台 从用 户那边把 原来大家 玩游戏的 时间给抢 走了。 很多人他 宁可去刷 短视频也 不喜欢游 戏了。 这是一个 原因,还 有一个是 整个行业 的新的玩法,

Like PUB G t here 's n o ne w ga mepl ay l ike that . In ter ms o f th e bu sine ss m odel Fr ee-t o-Pl ay i s pr obab ly t he b est busi ness mod el t hat we'v e fo und rece ntly bu t th ere' s no

new bus ines s mo del. The pro duct ion cost is gett ing high er a nd h ighe r. L ook at G TA, it's a 1 0-ye ar p rodu ctio n cy cle, it take s a lot of p eopl e to mak e so meth ing like tha t. A

nd t he c ost, the pri ce c an't be too high . Lo ok a t a game yo u're wat chin g a movi e he re, I es tima te t hat it's aro und $30- 40 p er h our. It can last you for 2 h ours 3 hour s. B ut s

ome game s ca n la st f or 2 0 ho urs. But it can only sel l fo r $3 0-40 . So thi s is a p rice cha llen ge. Ever yone is look ing at h ow t o se t th e pr ice for the late st v ersi on o f GT A. I

llen ge. Ever yone is look ing at h ow t o se t th e pr ice for the late st v ersi on o f GT A. I

f it can go to t he h ighe r pr ice rang e t hen othe rs w ill go t o th e hi gher pri ce r ange . At the sam e ti me, ther e ar e also Some big 3D mach ines lik e Wo ossi are sel ling at a ve ry l

. At the sam e ti me, ther e ar e also Some big 3D mach ines lik e Wo ossi are sel ling at a ve ry l ow p rice . So the ent ire indu stry is faci ng c hall enge s. T here are man y re ason s th at m

ake the gami ng i ndus try in a dow ntur n. T hey real ly w ant to g et m ore mone y. S o th ey n eed to m atch . So you rea lly star ted this pat h be caus e of gam ing? You sta rted to get inte

rest ed i n ga ming ? No I don' t ha ve m uch to d o wi th g amin g. I lik e to pla y ga mes and I li ke t o ma ke g ames . I' ve m ade some gam es m yself. But the reas on w hy I cho se t o do Mes

hie is n ot t he g ame itse lf. I ch ose to d o Me shie bec ause of some bas ic p rinc iple s. I

n th e se cond hal f of 202 2 I saw tha t Ch atGP T co uld gene rate lan guag e a nd S tabl eDif fusi on c ould gen erat e im ages . I saw that peo ple woul d go to a 3D mod el w ebsi te l ike Sket

chfa b to buy mod els. And the re w as n o pr oduc t in the mar ket that cou ld g ener ate 3D m odel s. S o we did it. Our idea is that sin ce e very one pays for 3D mode ls, if w e ca n ma ke 3 D mo

s. S o we did it. Our idea is that sin ce e very one pays for 3D mode ls, if w e ca n ma ke 3 D mo dels wit h a cost clo se t o ze ro, then we are rich . We are ric h no w. I t's okay . We sti ll m

ake 20% of o ur r even ue e very mon th. That 's f ast. 20% eve ry m onth . It 's d oubl ed i n th e pa st 1 2 mo nths . Wow. I wi ll p ay f or i t la st w eek. You sho uld tell me your ema il a ddre

ss. I wi ll g ive you a fr ee m embe rshi p. S o no w yo u ma inta in 2 0% g rowt h pe r mo nth. I'm

ss. I wi ll g ive you a fr ee m embe rshi p. S o no w yo u ma inta in 2 0% g rowt h pe r mo nth. I'm

from a f orei gn p ersp ecti ve, beca use I've tri ed y our text -to- text ure on t he w ebsi te.

What is the real bus ines s mo del of M esh? How is the busi ness ? Is it a go od b usin ess? It'

s ma inly fro m su bscr ipti on a nd A PI, thes e tw o ma in w ays. It' s ex actl y th e sa me a s Op enAI ? Ba sica lly, yes . Bu t we are act uall y a very act ive user com muni ty. So o ur c

onsu mer side has a v ery larg e nu mber of users. We h ave 2.5 mill ion user s pe r mo nth. And now

we have 4 m illi on u sers . Ho w bi g is the dif fere nce betw een this lev el a nd o ther 3DS L co mpan ies? How man y ar e ah ead of y ou? We h ave a to tal of 2 nd a nd 3 rd p lace . Th at's our

mpan ies? How man y ar e ah ead of y ou? We h ave a to tal of 2 nd a nd 3 rd p lace . Th at's our num ber of v isit ors. So you have mor e th an 5 0% o f th e ma rket ? In the US, we have abo ut 5

5% o f th e ma rket . An d th en, this is actu ally ... It's our mai n so urce of inco me. Subs crip

tion ? Ye s i t's main ly s ubsc ript ion. So ther e ar e al so 2 B AP Is? Yes, 2B APIs . An d th ere are also som e st udio s th at b uy s ome stud io a ccou nts. Stu dio game s? S tudi o ga mes.

To b e ho nest AI 3.0 is not real ly u p to the mar k ye t. T hey are stil l tr ying to fix it. They

are sti ll t ryin g to fix it. The y ar e st ill tryi ng t o fi x it . Th is i s wh at I did las t ni ght. I g ave him a pr ompt to crea te a goo d hu mano id a sset . An d th en D exte r's Hand

. De xter 's H and. Wow . Th at's a h igh requ irem ent. Tha t's a re sult for you ri ght? Yea h. Y

. De xter 's H and. Wow . Th at's a h igh requ irem ent. Tha t's a re sult for you ri ght? Yea h. Y

eah. So, in t he d atas et, ther e ma y no t be suc h da ta, but I fe el t hat it d oes gene rate a h and. But thi s ha nd w ill not have art icul atio n. I n th e fu ture we may do some art icul

atio n th ings . I' m go ing to p ay m ore. I n eed arti cula tion obj ects the mos t. N ow, many of our part ners Not only wil l th ey u se a rtic ulat ed o bjec ts, but they wil l al so u se M esh

to g ener ate a te a cu p a bow l a tab le, a ch air, etc . fo r th eir trai ning . Ho w bi g is the mar ket for 3D A I ge nera tion ? I thin k it 's o kay to d o a busi ness tha t ca n be lis ted. But

I'm not par ticu larl y sa tisf ied with thi s. B ecau se I thi nk o ur g oal shou ld b e to mak e a comp any like Inv itas. So, the core of our proj ect is t o ac cumu late mor e AI to do s omet hing

big ger than 3D. Thi s is wha t I' ve b een thin king abo ut. We h ave some pre limi nary att empt s. I thi nk M esh is g rowi ng v ery fast . Fo r me as lon g as I c an t hink abo ut w hat Mesh is

s. I thi nk M esh is g rowi ng v ery fast . Fo r me as lon g as I c an t hink abo ut w hat Mesh is befo re t he i ndex and the lin ear grow th a nd t he s low grow th, I ca n th ink abou t it. Do y

ou t hink 3D prin ting wil l ha ve a con sume r ma rket ? De fini tely . Ma ny u sers pri nt o ut t heir mod els usin g Me sh. Afte r pr inti ng, the cons umer mar ket will be bigg er t han the mark

et o f Ga meCu be. In t he U S y ou k now that 3D prin ters are the Kag er w ith a 20 % an nual exp ort volu me. They inc reas e by 20% eve ry y ear. Man y pe ople don 't h ave mode ls a fter the y bu

y a prin ter. So prin ting is also a v ery impo rtan t ma rket. But actu ally I can give a v ivid exa mple of this . Ou r th ree gene rati ons, act uall y m y ge nera tion and my pare nts' gen erat

ion are usin g AO M. A nd I 'm w illi ng t o pa y fo r it th ey'r e us ing it f or f ree. But I d on't thi nk I can con vinc e th em t o us e Me sh. Yes, def init ely. Thi s is act uall y a rela

tive ly s mall mar ket for the whol e 3D. For exam ple, if I se nd y ou a n OB J fi le o n We Chat yo u ca n't open it. Mos t pe ople don 't k now what GLB is. So its mark et s ize is j ust righ t fo

r a star tup to g row in. If y ou d o th is t hing too lar ge, for exam ple, vid eo m odel th is c hall enge is very big . It mea ns y ou c an't mak e it big . If you do this mar ket size is too

small, you can' t su ppor t th is c ompa ny. So a s a star tup, fin ding a m ediu m-si zed mark et c an a void com peti tion wit h gi ants and all ow y ou t o de velo p qu ickl y. Y ou j ust said you

r ul tima te g oal is t o ma ke a com pany lik e In vida . Ho w ca n Me sh b ecom e 10 0 ti mes, 100 0 ti mes bigg er t han it i s no w? I thi nk w e sh ould tak e it ste p by ste p. B efor e Hw ang

star ted Kuda he did n't expe ct t o re ach 40 m illi on a day. I th ink doin g bu sine ss i s co nsta ntly cha llen ging you rsel f c hang ing and subv erti ng y our own awar enes s c onst antl

y as king you r cu stom ers what the y ne ed t o so lve thes e pr oble ms, and then add ing some fir st-o rder thi nkin g. L ao H uang 's t hink ing is t hat this wor ld n eeds mor e co mput ing. If

I ha ve s omet hing tha t ca n gi ve y ou m ore comp utin g a nd i t's prog ramm able th en s omeo ne w ill try it o ut a nd s ee w hat prob lem it c an s olve. Mayb e at fir st i t wa s ju st a che

mist ry p rofe ssor who tri ed t o do a c hemi stry rea ctio n si mula tion . Bu t on e da y s omeo ne n amed Ale x Ka ndru bsky use d it to crea te a neu ral netw ork. And the n af ter 10 y ears we

rea lize d th at G enit y ca n ch ange the way hum ans live . Wh at a re y our firs t th ough ts o n th e de velo pmen t of Mesh? This is a go od q uest ion. I t hink wha t we are doi ng i s us ing

Mult i Mo del AI t o br ing peop le j oy. This joy is the joy of t he g ame. It' s th e jo y of 3D prin ting . Bu t if we go f urth er, it c an b e ot her form s of joy ev en n ot n eces sari ly 3

prin ting . Bu t if we go f urth er, it c an b e ot her form s of joy ev en n ot n eces sari ly 3 D mo dels . So I t hink if we c an s olve Mul ti M odel AI for fun, or more gen eral ly, AI f or f

un, then we can do a ver y bi g th ing. I ca n pe rsua de m y pa rent s to use it. Def init ely. You

can ask you r pa rent s to use it. Eve ryon e is int eres ted in t he A I fo r fu n. A nd e very one is i ts u ser. Her e's a go od e xamp le. I up load ed a pic ture of me a nd m y gi rlfr iend yes

terd ay. And I cr eate d an ass et. And then th e fi rst year I s ent it t o my gir lfri end, she

terd ay. And I cr eate d an ass et. And then th e fi rst year I s ent it t o my gir lfri end, she sai d it was ugl y. B ut I bel ieve if you can make it look goo d s he w ill be w illi ng t o 3D pri

nt it. Righ t t his is a ctua lly a di rect ion we a re t akin g. B ecau se t he u se o f th e ph

nt it. Righ t t his is a ctua lly a di rect ion we a re t akin g. B ecau se t he u se o f th e ph oto scen e is not our mai n us e sc ene. Gen eral ly s peak ing, we are more foc used on the game or

a si ngle mon ster or prot agon ist, and we want to prin t it out . Fo r th e hu man side we sti ll h aven 't d one a pa rtic ular ly g ood job. Eve ryon e wi ll t hink tha t yo u ma de t his. Rig

ht. Ever yone is a Me sh u ser. Yes bu t ac tual ly M esh' s gr owth is very suc cess ful. It has

grow n 12 tim es a yea r. A lmos t. B efor e th is, the busi ness of Tai Chi was not so s moot h. I

act uall y wa nt t o he ar m ore abou t yo ur u p an d do wn a s a PSD. I w ant to h ear more abo ut y our up a nd d own as a CEO . I thin k I can face thi s now. When you are in the low end, you

wil l th ink why are you so s tupi d? Y ou c an't do anyt hing . No w y ou h ave done a b usin ess to t he p oint of prod uct mark et f it. And ther e is rap id g rowt h. A nd i n th is i ndus try,

fro m th e pe rspe ctiv e of mar ket shar e i t is the fir st p rodu ct i n th e fi rst plac e. S

o I thin k I can face the pas t mo re f rank ly. I ca n fa ce t he p robl ems I ha d be fore . Ba

sica lly, the fir st 2 4 mo nths, 18 m onth s to 24 mont hs, abou t a year and a h alf to t wo y ears we wer e al l th inki ng a bout wha t ki nd o f pr oduc t we sho uld make . An d th en t here was no. .. T here was no sign of any

comm erci aliz atio n. B efor e yo u th ough t ab out it, you had alre ady deci ded to s tart a s tart up. Yes. So you didn 't c are abou t it at that tim e y ou d ecid ed t o st art a st artu p fi

tart up. Yes. So you didn 't c are abou t it at that tim e y ou d ecid ed t o st art a st artu p fi rst. Yes I thin k th is i s so meth ing that I a nd o ther s ar e no t th e sa me. I am ver y bi

rst. Yes I thin k th is i s so meth ing that I a nd o ther s ar e no t th e sa me. I am ver y bi ased for act ion. If ther e is som ethi ng I wan t to do, I w ill do i t im medi atel y. I wil l do

it even if I do n't slee p at nig ht. And when I o ften hav e to do some thin g I thi nk a bout it at n ight . Ma ybe I ha ven' t fi nish ed t he 8 -cha ract er. It's a b it l ike Elon Mus k's rock

et. The rock et i s no t re ady yet, and he' s th inki ng a bout wha t we 're goin g to eat on Mars

et. The rock et i s no t re ady yet, and he' s th inki ng a bout wha t we 're goin g to eat on Mars . I thin k th is i s on e of my char acte rist ics. Man y pe ople wil l th ink, "Ar e yo u th inki

. I thin k th is i s on e of my char acte rist ics. Man y pe ople wil l th ink, "Ar e yo u th inki ng t oo f ar?" Or "I h aven 't f inis hed the 8-ch arac ter, how can I t hink of thes e th ings ?" B

ut I wil l th ink abou t it in the midd le. - Mm -hmm. We d id a lot of inte rest ing thin gs, but

we a lso trie d so me c omme rcia liza tion met hods . We fou nd t hat comm erci aliz atio n ca n on ly b e do ne b y ou tsou rcin g. O utso urci ng m eans ... We t ake a pr ojec t an d th en w e de

velo p it for som eone els e. W e de velo p it for him and the n we han d it ove r to him . Th is i s a very low -cos t jo b. A nd i n Ch ina, suc h a thin g is bas ical ly i mpos sibl e to do. So

we d ecid ed t o sw itch to In t his proc ess, I r eali zed that I c an o nly do s tand ardi zed face -to- face con sume r pr oduc ts. Why is t his rela ted to y our pers onal ity? Bec ause I h ate

to d eal with big cli ents . So you 'd r athe r do API and sub scri ptio n i nste ad o f ta lkin g ab out the exte rnal pac kage . Or at leas t st art with thi s. O f co urse it can be done in the

futu re. Of c ours e I wou ld b e wi llin g to do busi ness wit h it . Bu t at the ear lies t st

futu re. Of c ours e I wou ld b e wi llin g to do busi ness wit h it . Bu t at the ear lies t st age, I t hink it' s al read y su ch a dif ficu lt t hing to figu re o ut p rodu ct m arke t fi t. S

o d on't add mor e de tail s to mak e yo urse lf d o so meth ing you' re n ot g ood at. Look ing

back at the two year s be fore Tai ji, thin gs w eren 't g oing wel l. Y ou c an n ow f ace it f ace to f ace. Wha t mi stak es d id y ou m ake? Did you fee l th at y ou w ould nev er b e ab le t o do

it agai n? A ctua lly, fro m no w on I thin k ab out what I d id i n Ta iji two year s ag o. I did

n't get any resu lts. It' s ju st t hat the busi ness sit uati on i s no t su stai nabl e. I thi

nk f or a sta rtup th e mo st i mpor tant thi ng i s to fin d a a ma ture mon etiz atio n or a m ode of m akin g mo ney, so that you can run it with a s elf- suff icie ncy. The big gest pri ncip le i

n th is i s th at t he t echn olog y or pro duct you mak e mu st b e co mmer cial ized . If a t echn olog y ca nnot be comm erci aliz ed, you cann ot u se c omme rcia liza tion to get reve nue to s

uppo rt f utur e te chno logy res earc h. T his tech nolo gy i tsel f wi ll b ecom e a fire work s sh ow. Afte r th e fi rewo rks show it wil l be gone. But you will wan t to tur n th e te chno

logy dev elop ment int o a mark et. Yes, int o a flyw heel . Th at i s y ou d o go od t echn olog y an d ge t go od p rodu cts. Goo d pr oduc ts a nd g ood reve nue. Goo d re venu e ca n at trac

t go od p eopl e to do good tec hnol ogy. Thi s wa y y ou c an k eep maki ng m oney . Th is w ay, you can make the com pany big ger and bigg er. But it's qui te d iffi cult to figu re o ut s uch

a pa tter n. A nd e spec iall y wh en y ou a re n ot s ure at m any time s. I thi nk i n 20 21, ther e ar e a few term s th at w e ma y no t me ntio n no w. F or e xamp le, the orig inal uni verse. For

exam ple, ope n so urce in fras truc ture . Th ese word s we re v ery popu lar at t he t ime. At

that tim e y ou w ould giv e up you r th inki ng a bout som e th ings . Yo u wo uld list en t o wh at t he t rend of this soc iety is talk ing abou t. A nd t hen you woul d th ink abou t wh at y ou c

an d o or not . Yo u wo uld thin k o h i s wh at I do usef ul i n th is? That 's a ctua lly a ve ry d iffi cult thi ng t o av oid. But whe n yo u ha ve m ore busi ness exp erie nce, you may not ste

p in to s ome basi c ho les. So f rom Taij i to the piv ot o f Me sh, I wa nt t o bo rrow you r ar ticl e. Y ou t alke d ab out, it shou ld b e In tel. Yes . DR AM b usin ess. Oh, Gol denM oor. Yes

ticl e. Y ou t alke d ab out, it shou ld b e In tel. Yes . DR AM b usin ess. Oh, Gol denM oor. Yes

. Th is c onve rsat ion shou ld h ave happ ened nex t to Int el's Off ice. Can you tel l us abo ut t his stor y ag ain? Yes . Th at w as a roun d 19 70. At t hat time th e en tire sem icon duct or i

ndus try in t he U nite d St ates mai nly made DRA M. I t fa ced a lo t of cha llen ges. Bec ause in Japan, The Japa nese eng inee rs w ere very dil igen t an d th e se rvic e wa s go od. They wor ked

over time and wor ked with sev eral com pani es. Basi call y I ntel 's D RAM coul dn't sel l. I t co uldn 't b eat the Japa nese pro duct . At tha t ti me, Gord on M oore th e fo unde r of the Moo re L

aw, and anot her foun der of I ntel wa s... It shou ld b e An dy G rove . An dy G rove. Andy Gro

ve w rote a b ook, "On ly t he g eniu ses surv ive. " An d "T he H igh- Scho ol M anag emen t of the Hig h-Pr ivat e" a re a ll g ood book s. A ndy Grov e an d hi s wi fe w ere talk ing in t he o ffic

e. " What sho uld we d o? W e ca n't cont inue thi s co mpan y." "Our DRA M is no long er a ble to b

e. " What sho uld we d o? W e ca n't cont inue thi s co mpan y." "Our DRA M is no long er a ble to b eat the Japa nese ." A ndy Grov e sa id, "If we c onti nue this th e bo ard will fir e us ." " You

know th e bo ard in t he U S ha s a lot of p ower . Th ey c an c hang e th e CE O." Andy Gro ve s aid, "If we cont inue thi s t he b oard wil l fi re u s." Why don' t we fir e ou rsel ves and leav

e th is r oom and come bac k? T hen we t houg ht, if w e we re t he n ew C EO o f th is c ompa ny, what wou ld w e do ? Th ey c ame to a con clus ion that if this com pany hir es a noth er n ew C EO,

they wou ld g o an d do CPU . Th is i s th eir stor y. T his is t heir sto ry. Late r t hey spen t th ree year s to tra nsfo rm f rom DRAM to CPU. It was very ve ry p ainf ul. But this als o cr

eate d th e gl ory of I ntel for dec ades. Was it a pro cess fro m Ta iji to M esh that Hu Yuan ming sta rted ? Ye s I thi nk s o. H ow w as y our thin king pro cess ? Wa s it pai nful ? I thin k it

sta rted ? Ye s I thi nk s o. H ow w as y our thin king pro cess ? Wa s it pai nful ? I thin k it was a l ittl e bi t. B ecau se i t wa s li ke.. . I want ed t o pr ove myse lf a s a prog ramm er t

o ma ke a suc cess ful comm erci aliz ed p rodu ct. I ju st g ave up. Afte r al l T aiji has bee n do ing this for a l ong time . I just did n't do it. But I re aliz ed t hat if I con tinu ed t

o do Tai Chi I woul d no t se e th e fu ture of busi ness . At lea st I can 't s ee i t as a c ompa ny-o rien ted busi ness . It 's l ike I'm firi ng m ysel f. I n fa ct, I ha ve t o fi re m ysel f ev

ery 3 to 6 m onth s. E very 3 t o 6 mont hs, I ha ve t o be come a w hole new me. You hav e to rei nven t yo urse lf e very 3 t o 6 mont hs. If y ou w ant to s urvi ve i n su ch a n in tens e en viro

nmen t of ent repr eneu rshi p i t's best to thin k th at y ou w ere a fo ol 3 mon ths ago. This is

the best . If you don 't t hink you wer e a fool thr ee m onth s ag o i t me ans you have n't grow n in the las t th ree mont hs. This is diff eren t in sch ool. In scho ol, beca use ever yone is

well pro tect ed b y th e sc hool it doe sn't mea n th at t hey are bank rupt or expe lled by the teac her or h ave a li fe-s avin g pr essu re o r so meth ing. In scho ol, it's an open env iron ment wit hout lif e-sa ving pre ssur e. B ut t he s tart up c ompa ny i s co mple tely dif fere nt.

I ba sica lly turn ed m y mi ndse t fr om a aca demi c mi ndse t in to a bus ines s mi ndse t. O r I took the two I thin k b ette r a nd i nteg rate d th em t oget her to m ake the comp any more com

peti tive . At tha t ti me, you and the enti re T ai G roup sai d " We w on't do this any more . We

peti tive . At tha t ti me, you and the enti re T ai G roup sai d " We w on't do this any more . We

'll do t he M anag ing Dire ctor th e 3D Ass et B usin ess. Do you have our sup port ?" T here was som e re sist ance . So me p eopl e al so l eft the comp any beca use they wer e no t in tere sted

in this. But thos e wh o in sist on doin g it now thi nk t his is a ver y co rrec t de cisi on. Of c

in this. But thos e wh o in sist on doin g it now thi nk t his is a ver y co rrec t de cisi on. Of c

ours e m any stud ents who do Tai Chi stil l st ay i n th e co mpan y. R ecen tly, we were dis cuss ing how to d o Ta i Ch i 2. 0. W e th ink ther e ar e st ill a lo t of thi ngs to d o. T his is s

till ali ve. At t hat time ac tual ly, ever yone was con cern ed a bout 3D. Act uall y w e st arte d do ing 3D b ecau se o f Ta i Ch i. B ecau se a t th e ea rlie st, ever yone use d Ta i Ch i to ach

ieve hig h-en d sp latt ing. The n th ere were als o st uden ts w ho u sed Tai Chi to a chie ve n erf. Then we thou ght, "Oh Ta ikin can do this ." C an w e pu sh t his thin g to ward s th e di rect

erf. Then we thou ght, "Oh Ta ikin can do this ." C an w e pu sh t his thin g to ward s th e di rect ion of r eal esta te? The earl iest rea l es tate can be draw n. T hen, gra dual ly, in s ome scen

ario s i t ca n be don e re lati vely well. At t hat time th ere was a lo t of res ista nce. And we

also had ano ther pro ject . We lat er c ut a noth er p roje ct a nd m ade it a ll-i n to do Mesh . Bu t in fac t M esh is n ot r elat ed t o th e te chni cal barr iers you bui lt o n Ta iji. A p

. Bu t in fac t M esh is n ot r elat ed t o th e te chni cal barr iers you bui lt o n Ta iji. A p

erso n wh o is par ticu larl y go od a t wr itin g Ta iji may not be a ble to p ut 3 D as set gene rati on v ery well . Te xt t o te xtur e e tc. It's lik e yo u st arte d fr om s crat ch. Yes, we

are comp lete ly a tea m th at d oesn 't u nder stan d to o mu ch a bout AI. No one in t he t eam has ever use d a big mode l. W e ar e a grou p of peo ple in g raph ics You know gr aphi cs p eopl

e ha ve a big cha ract eris tic, whi ch i s th at t hey are very har d-co re. They can lea rn a nyth ing. Bec ause if you play Ola GaGa eve ry d ay, and toda y yo u le arn a di ffus ion, tha t's too

ing. Bec ause if you play Ola GaGa eve ry d ay, and toda y yo u le arn a di ffus ion, tha t's too easy . If you pla y MC MC a ll d ay, toda y yo u le arn SDE, the se a re a ll t he s ame. So in f

easy . If you pla y MC MC a ll d ay, toda y yo u le arn SDE, the se a re a ll t he s ame. So in f act, a l ot o f gr aphi cs s tude nts are now very eas y to get by with AI. I k now a lo t of stu

dent s li ke t his. Grap hics peo ple are very har d-wo rkin g. T hey lear n th ings ver y qu ickl y. S o ev eryo ne c an q uick ly s tart doi ng a lot of thin gs i n it . Wh en y ou w ere doin g Me

y. S o ev eryo ne c an q uick ly s tart doi ng a lot of thin gs i n it . Wh en y ou w ere doin g Me sh, your und erly ing logi c y ou w eren 't w orri ed t hat your tea m wa sn't sui tabl e. W e ha

ven' t do ne t his kind of text -to- imag e t ext- to-3 D as set. Wou ld y ou w orry abo ut t his?

Wou ld y ou w orry tha t yo ur t eam is n ot c ompe titi ve e noug h? I was n't worr ied abou t th at. Beca use no o ne w as d oing thi s at the tim e. O h n o on e wa s do ing this at the time . We

at. Beca use no o ne w as d oing thi s at the tim e. O h n o on e wa s do ing this at the time . We

wer e th e fi rst onli ne p rodu ct t hat coul d be pub licl y vi site d. T he p rodu ctio n wa s ve

ry p oor. Ha h a ha ha ha ha I fa iled bu t it 's o kay. How did you det ermi ne t hat the mark et c

ould exi st? If n o on e di d it it 's h ard to p redi ct h ow b ig t he m arke t wo uld be. How did

you keep tha t in min d? Y ou s houl d ha ve o ther piv otin g di rect ions . Wh y Me sh? To b e ho nest Me sh w as t he b est pivo ting dir ecti on. I di d LL M 2 D di ffus ion, tex t-to -ima ge,

or t ext- to-v ideo . I coul d ne ver do i t be tter tha n ot hers. My a bili ty t o li mit myse lf l ed t o th e fa ct t hat Mesh was my best dir ecti on a t th e ti me. This is also the adv anta

ge o f th e en tire tea m's abil ity. So you clea rly reco gniz ed t he b ound ary betw een your abi lity and the tea m's abil ity. Yes . Bu ild some thin g th at t he w orld nee ds a nd b uild som

ethi ng t hat you want . At tha t ti me, you thou ght this was wha t th e wo rld need ed. Yes. How

det ermi ned were you ? I was very det ermi ned beca use alth ough the pre viou s pr oduc t fa iled we tal ked to a lot of user s an d th ey a ll s aid, I wo n't pay for your sof twar e. B ut i f yo

u se ll m e th e mo del insi de, I wi ll p ay f or i t. W e ta lked abo ut a noth er t hing tod ay.

You make a 3 D as set of a per son, and I c an c onvi nce my p aren ts t o pa y fo r it . Ex actl

y. S o a lot of time s w hen you talk to your use rs, you can get new solu tion s. S o y ou a sked me how big the mark et s pace of Mesh is. I t hink it' s in fini tely big for a c ompa ny l ike us t

hat has a st rong ada ptab ilit y. B ecau se y ou c an a lway s so lve new prob lems. You can alwa ys s olve mar ket- need ing prob lems thr ough a g roup of smar t pe ople an d ma ke a pro duct wit

h co mpet itiv enes s a nd t hen divi de i t am ong a ne w ma rket . Yo u sa id i n yo ur i nter view in 2013 tha t th e te chno logy of the AI-b ased 3D mode l on ly s olve d 10 % of the pro blem

at that tim e. Y ou s aid that you mig ht b e ab le t o so lve 90% in 2 025. Now tha t 20 25 i s he re, how much do you thin k yo u ha ve s olved? Very goo d qu esti on. Mayb e 50 %? 5 0%? Yes. Bec

ause the mor e yo u so lve it, the more you fin d ou t th at t he t op l ine is n ot 1 00%. So I th ink now it m ay s olve 90% of what I i magi ned at t he t ime. But lat er I fou nd o ut t hat the

top line is 200% . Is the rem aini ng 5 0% s olva ble? In the futu re t hat can be s een? The rem

aini ng 5 0% I thi nk i s so lvab le. I th ink the bigg est thin g I lear ned in t his proc ess is n ot t o un dere stim ate the spee d of tec hnol ogy. When we firs t ma de M esh, we used ver y ol

d an d br oken tec hnol ogy. But we kept on usin g it . No w w e ca n so lve many pro blem s th at a re m uch bigg er t han befo re. We c ould onl y us e it for hor ror game s. W hy h orro r ga mes?

Bec ause we crea ted four fac es f or a ll t he m odel s. W hy f our face s? B ecau se w e cr eate d th e mo dels an d th en w e cr eate d a figu re f or e ach mode l. S o yo u ca n se e th e ch arac

ter' s he ad, one face in the fron t o ne f ace in t he s ide, and one fac e in the bac k. T here

are fou r fa ces in t otal. Only hor ror game s ca n us e th is k ind of a sset . Bu t no w we can see som e do uble -age d st udio ev en s ome trip le-a ged stud io, usin g Me sh t o ma ke s ome

prot otyp e. S o wi th t he c onti nuou s te chno logy it has a v ery, ver y la rge spac e. A nd I thi nk e ven toda y w e ca n on ly d eal with the 5% shar e of the mar ket. The re i s st ill a lo

t of spa ce b ehin d. F rom the succ essf ul c omme rcia lize d Me sh t o th e pr esent, the pres sure of Tai Chi. How did you ove rcom e th at? Beca use you went fro m be ing a br ight and bri ght

pers on t o be ing a bu sine ss p erso n of Tai Chi . Ev eryo ne t houg ht y ou w ere not doin g we ll. How did you deal wit h th at? Did you feel the pre ssur e an d fr ustr atio n? O f co urse I d

ll. How did you deal wit h th at? Did you feel the pre ssur e an d fr ustr atio n? O f co urse I d id. But fran kly spea king wh en T ai C hi w as n ot g oing wel l I did n't feel muc h pr essu re.

id. But fran kly spea king wh en T ai C hi w as n ot g oing wel l I did n't feel muc h pr essu re.

I th ink the pres sure now is simi lar to t he p ress ure befo re. But I th ink I wa s mo re c onfi dent in my b usin ess. Whe n I was work ing at T aiji I didn 't e ven know if I wa nted to do b usin

ess. You did n't run to m ake mone y? I did n't run to m ake mone y at the ear lies t. I f I had,

ess. You did n't run to m ake mone y? I did n't run to m ake mone y at the ear lies t. I f I had, I w ould hav e us ed a noth er w ay. We t ried to do b usin ess with oth er p rodu cts, but non e of

the m wo rked . So Me sh w as o ur t hird pro duct . At tha t ti me, a lo t of pre ssur e ca me f rom I ha d no ide a ho w to run a c ompa ny. Befo re I bec ame a CE O I onl y ha d a PhD. I h ad n

ever wor ked anyw here . Be fore I b ecam e a CEO, I w as a n in tern . My mos t pr ofes sion al e xper ienc e wa s at NVI DIA, Ado be, and Micr osof t. T his is m y kn owle dge of t he f ield bef

ore I be came a C EO. So, when I f irst sta rted I didn 't t hink abo ut h ow t o be come a C EO.

I ju st t houg ht a bout wha t a norm al c ompa ny s houl d be lik e. I did n't know. And then I l earn ed s lowl y. A nd t oday I'm sti ll a CEO who is alwa ys s tron g. S o in the pro cess th e ea

rlie st s trug gle is o ften her e. Y ou d on't kno w ho w to run a c ompa ny. Yes, I d on't kno w ho w to run a c ompa ny. And then our fin anci ng w as a lso very smo oth. The re w as n o pr essu

re t hat I wo uld die if I did n't get paid in a mo nth. So in f act, thi s th ing dela yed me f rom

faci ng s ome crue l fa cts. I But I ta lked to some of my f ound ers rece ntly an d I was also sur

pris ed. Beca use I fo und that som e of my olde r 20 -yea r-ol d fo unde rs c ame out of t he b

pris ed. Beca use I fo und that som e of my olde r 20 -yea r-ol d fo unde rs c ame out of t he b ig f acto ry t o st art a co mpan y. I t se ems that the y di d it bet ter than us. So I th ink this

is not beca use of m e. I t's not beca use I wa s an int ern befo re I bec ame a CE O. S o I have thi s ch alle nge. Thi s is a c ommo n ch alle nge that eve ryon e ha s. C an a goo d CE O be tra

ined ? I thin k so . I thin k a good CEO ... The bigg est adva ntag e is to be a ble to l earn qui

ined ? I thin k so . I thin k a good CEO ... The bigg est adva ntag e is to be a ble to l earn qui ckly and imp rove you r kn owle dge. So alth ough I m ade a lo t of mis take s I hav e a adva ntag

e in tha t I chan ged myse lf v ery quic kly. You jus t ta lked abo ut t he C EO w ho c ame out of a lot of big comp anie s. I saw a l ot o f id eal spee ch. He e ven came out of high sch ool afte r hi

gh s choo l. W hen you meet thi s ki nd o f th ing, you wil l th ink, OK, why is it t hat idea l di dn't fin ish high sch ool and he d id b ette r th an u s in the com pany ? Th is i s al so w hat you

ment ione d in tha t ar ticle. Why? Bec ause PhD is not impo rtan t. T hat' s wh at I mea n. I thi

nk i t's help ful to d o so meth ing for your self . Bu t do you nee d a PhD to d o it wel l? N ot n eces sari ly. Ther e ar e ma ny t hing s th at a PhD stu dent can lea rn i n th is p roce ss. For

exam ple, hum ble, wor k ha rd, pres enta tion ski lls. You can exp lain thi ngs clea rly. And you

can ... know som e ad vanc ed t echn olog y. B ut t here are man y th ings tha t a PhD can' t ge t to . Fo r ex ampl e t he m arke t u sers ma ss p rodu ctio n of pro duct s. H e ca n't get to t

t to . Fo r ex ampl e t he m arke t u sers ma ss p rodu ctio n of pro duct s. H e ca n't get to t hese thi ngs. For exa mple Li Xia ng. He i s a very goo d en trep rene ur. He i s al so v ery stra

ight forw ard. Whe n he see s so me u nple asan t th ings he wil l sc old him. He also exp erie nced ... When he used to work for bub ble web, he left wor k wi thin a d ay. He h as e xper ienc ed m any

... When he used to work for bub ble web, he left wor k wi thin a d ay. He h as e xper ienc ed m any thin gs. So d urin g th is p roce ss, he w as v ery humb le a nd l earn ed. He e ven foun d a CEO

thin gs. So d urin g th is p roce ss, he w as v ery humb le a nd l earn ed. He e ven foun d a CEO to r epor t to the CEO and tak e ov er t he p rodu ct. You can see that thi s pa tter n is the mos

t im port ant thin g. H ow b ig a CEO has a p atte rn, how big a th ing can he d o. S o fa r d o yo u th ink you are a go od C EO? I th ink it d epen ds o n wh o yo u co mpar e wi th. If y ou s ay t

hat the grow th o f Me sh i s fa st, then you can 't b e fa st c ompa red to L ovab le a nd C ursor.

But comp ared to trad itio nal SaaS tr iple tr iple do uble do uble do uble tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple

tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr

iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple

tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr

iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple

tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple tr iple, and be m

ore deci sive in some mat ters . Fo r ex ampl e l earn fas ter. Spe nd m ore time out side no t in side . Fo r ex ampl e r ead more boo ks. Refl ect more . Th ink abou t th e co mpan y's visi on.

side . Fo r ex ampl e r ead more boo ks. Refl ect more . Th ink abou t th e co mpan y's visi on.

Draw a b luep rint . Fo r ex ampl e r ecru it. I th ink ther e ar e ma ny t hing s to imp rove . Wh

at l esso ns d id y ou l earn as a CEO? You' ve p laye d th e ga me 1 2 ti mes, fro m 48 hou rs a go, to n ow. Whic h mo ment was the mos t di ffic ult thin g yo u le arne d? T he m ost diff icul

t th ing you lear ned was then den ied by y ours elf. The mos t di ffic ult thin g I lear ned was then den ied by m ysel f. T here wer e ma ny t imes . Le t me thi nk a bout the big gest one . I

thin k th e bi gges t on e is tha t I real ly h oped tha t ev eryo ne w ould lik e me . I hope d th at I wou ld b ecome Ever yone tho ught tha t Yu emin g wa s a nice per son, a g ener ous pers on,

and a go od f rien d. B ut a fter a l ot o f th ings I real ized tha t if you r go al i s to mak e ev eryo ne l ike you, you wil l ev entu ally be hate d by eve ryon e. S o I' m no t pu rsui ng e

very one to l ike me, but I'm purs uing to make imp orta nt p eopl e li ke y ou. Peop le w ho d on't car e wi ll s cold you or thin k yo u're a g ood pers on. As l ong as y ou t hink you 're doin g th

e ri ght thin g i t's fine . Th ere will alw ays be p eopl e wh o wi ll s cold you . I thin k an othe r th ing is t o lo ok a t th e wh ole team fro m th e pe rspe ctiv e of a m anag er. Is a goo

d le ader a l eade r th at e very one like s? N ot n eces sari ly. A go od l eade r ma y be a l eade r th at e very one is h appy wit h b ut i t ma y no t ne cess aril y be a l eade r th at e very one

likes. But he s till wan ts t o fo llow you . Wh y? B ecau se h e ca n ga in g rowt h he re. He c

likes. But he s till wan ts t o fo llow you . Wh y? B ecau se h e ca n ga in g rowt h he re. He c

an d o wh at h e wa nts to d o he re. As f or w heth er e very one like s yo u I thi nk i f yo u pu t th is a s yo ur f irst pri orit y m any time s it wil l le ad t o th e wr ong deci sion . Yo u do

n't care if othe rs l ike you as y our firs t pr iori ty. What is your fir st p rior ity? To succ

eed in t hing s. W in. My f irst pri orit y is to succ eed in t hing s. A nd I rea lize d th at y ou h ave to m ake a... you have to do s omet hing goo d t hen some one will giv e yo u a nega tive

eva luat ion. You mea n th at i n th is p roce ss, you have to be i nevi tabl e y ou h ave to b e de live rabl e fo r th e sa ke o f pr ogre ss, you have to make tou gh d ecis ions th en s omeo ne w

ill hate you . Bu t th at's fin e. W hat is a tou gh d ecis ion? For exa mple we hav e to cut off a p roje ct t hat ever yone lik es. For exam ple, you may hav e to inv ite a co llea gue who has

no c ontr ibut ion to t he t eam, but eve ryon e li kes him, he is a goo d pe rson. For exam ple, whe n yo u ar e ma king a p rodu ct, ther e ar e A and B us ers. If you make A l ike it, B wi ll s

cold you . If you mak e B like it, A w ill scol d yo u. D o it all . A and B do it all. The com pany

cold you . If you mak e B like it, A w ill scol d yo u. D o it all . A and B do it all. The com pany is done . Th ey w ill all scol d yo u. T hey will all sco ld y ou. So I thi nk t his is a bar rier

is done . Th ey w ill all scol d yo u. T hey will all sco ld y ou. So I thi nk t his is a bar rier tha t ev eryo ne h as t o cr oss. It' s ca lled bec omin g an adu lt. The stud ents in scho ol m

ay n ot b e ab le t o st and it. How can othe rs c riti cize me? But onc e yo u cr oss it, you will fin d th at t he w orld is very wid e. I f so meon e sc olds me, jus t sc old me. And you lear ned

this aft er y ou b ecam e th e CE O? Y es, beca use in s choo l e very one is v ery nice . Ev en a t MI T e very one is l ike, "He llo, hel lo, hell o." And they avo id a lot of conf lict s. A nd n

ow y ou w on't avo id t hem? I w on't avo id t hese con flic ts. As l ong as t he p roje ct i s su cces sful th ere will be conf lict s. Y es, beca use only in this way ev eryo ne i n th e te am i

s re spon sibl e. I f yo u av oid thes e co nfli cts, you are a s elfi sh p erso n. B ecau se b asic ally yo u wa nt to by a sma ll g roup of peop le, but you are not resp onsi ble for ever yone 's i

nter ests . Yo u ar e in the pos itio n of CEO yo u ha ve t o be res pons ible for the ent ire comp

nter ests . Yo u ar e in the pos itio n of CEO yo u ha ve t o be res pons ible for the ent ire comp any' s in tere sts, ins tead of lett ing your self be like d by eve ryon e. I am like d by eve ryon

e be caus e of my own busi ness it is priv ate. The suc cess of the comp any is t he c ompa ny's bus ines s i t is eve ryon e's busi ness pr ivat e ma tter s ar e be hind the com pany . La ter I fo und

that bas ical ly a ll t he m anag ers, I h ave two laye rs b elow me, I o ften enc oura ge s ome peop le t o be man ager s. M any mana gers wh en t hey firs t be come man ager s t hey all have thi

s pr oblem. It's too har d to say . Ye s. W hy i s th ere such a p robl em? Beca use when the y we re d oing IC, ind ivid ual cont ribu tors th ey w ere very str ong ICs. So thei r ma nage rs d idn'

t ne ed t o pu sh t hem. The y di dn't nee d to do anyt hing bad to them . Bu t wh en t hey were pro mote d to the man agem ent posi tion th ey t houg ht t hat the way they wer e tr eate d as ICs was

the way the y we re t reat ing thei r te am. But here com es t he q uest ion. The se p eopl e yo u're wor king wit h no w us ed t o be you r pe ers. The rea son you were pro mote d is bec ause you

r se lf-d isci plin e an d yo ur p robl em-s olvi ng a bili ty a re b ette r th an t heir s. S o yo u ha ve t o gi ve t hem high er s tand ards . Bu t ma ny n ew m anag ers can' t gi ve t hem high

er s tand ards . Th ey w ant to b e li ked by t heir own tea m. S o yo u've mad e a chan ge. How

do y ou f orce you r ma nage r to mak e th e sa me c hang e? T ell them abo ut t he p ast. Tel l th em w hy t he c ompa ny h as t his prob lem toda y. I t's beca use I wa s a bit of a mes s ha lf a

yea r ag o. I wan ted ever yone to like me. I wa nt m y fr iend s to thi nk t hat I am a p erfe ct p erso n w hich led to the comp any' s pr oble m to day. I t hink I s ee a lot of thes e ch arac teri

stic s in Elo n wh en y ou t alk abou t th e so fa. He's too ext reme . He 's l ike, you can 't d o it tod ay, but you' ll l eave the com pany in an h our. I t hink we' re s till a r elat ivel y to

lera nt p lace tha t al lows eve ryon e to mak e mi stak es. But I th ink Elon has a l ot o f go od p oint s. H e's a go od e xamp le i n ma ny p laces. I he ard an i deal ist' s co mmen t on Elo

n. H e sa id h e wo uld like to eval uate a C EO. Not to s ee h ow f lexi ble he i s h ow m uch he c

n. H e sa id h e wo uld like to eval uate a C EO. Not to s ee h ow f lexi ble he i s h ow m uch he c an d o th ings ho w mu ch h e ca n ma ke g ood busi ness . Bu t to see if a CE O ca n se e th e es

senc e of the pro blem at the real cri sis. He said you can see Elo n ma king so many thi ngs that peo ple don' t li ke. He m ade so m any thin gs t hat peop le d on't lik e. B ut a t ev ery impo

rtan t mo ment of Tesl a Sp aceX he can see the ess ence . He thi nks this is the only thi ng t he C EO s houl d do. Yes, a g ood CEO has only one tas k w hich is to l ead the comp any to w in. The

rest is not impo rtan t. E ven if a lot of peop le h ate you? If that 's n eces sary th en a lot of peop le h ate Elon . To o ma ny p eopl e. A ll o f Ji nson 's d irec t re port s fe el t hat he i

s ve ry h ard to w ork with . Do es i t af fect him bei ng r ated as the most pop ular CEO ? No it

doe sn't . Wh y? B ecau se m ost of t he N VIDI A pe ople don 't r epor t to him . So the y do n't have to be t ortu red by J inso n ev ery day. But Jin son did a go od j ob. Ever yone is sati sfied.

You ment ione d th at y ou a re a goo d CE O b ut y ou c ompa re w ith othe rs. You ment ione d th at y ou a re a com pany lik e Cu rsor . Do you thi nk t hat Curs or's bus ines s is too goo d? A

ctua lly, you r ef fort s ar e th e sa me a s th eirs . Yo u ar e ha rdwo rkin g a nd t hey are hard work ing. But the y ea rn m ore mone y th an M essi . Ye s. S o a s a CEO, the mos t im port ant

work ing. But the y ea rn m ore mone y th an M essi . Ye s. S o a s a CEO, the mos t im port ant thin g is to choo se a goo d tr ack. If you choo se a goo d tr ack, of cour se, ther e ar e fa ctor

s of luc k. Y ou c an m ake mone y in tha t fi eld. You are ... If y ou w ant a CE O wh o is luc ky a nd c an b et, or a CEO who is good at work ing but can only bet mo st p eopl e wi ll w ant a CE

O wh o is luc ky a nd c an b et. But I ca n on ly s ay t hat Cust er's dev elop ment is also tem pora ry. If h is d evel opme nt t his time is beca use he c hose thi s tr ack, if it's luc k t hen

pora ry. If h is d evel opme nt t his time is beca use he c hose thi s tr ack, if it's luc k t hen you will hav e to ret urn the luck you got wit h yo ur s tren gth soon er o r la ter. For exa mple

he has 200 mil lion AR now. What kin d of bus iness? Is i t a good bus ines s? W hat are its char

acte rist ics? Is Curs or a goo d bu sine ss i n yo ur o pini on? I th ink the mark et i s bi g. T

he m arke t is big and it' s no t a comm on m arke t. T his is a ctua lly very dif ficu lt. You have

to see a ma rket whe n yo u do n't beli eve ther e is a m arke t he re. In f act, Mes h is the ear lies t th ing all 3D A I pl ayer s do now. We s tart ed d oing 3D mode ling whe n we wer e no n-co

mpet itiv e. E very one thou ght it w as a sma ll m arke t. B ut w hat we c an d o no w is a m ediu m-si zed AI c ompa ny. The mark et i s bi g an d no n-co mpet itiv e. A nd y ou h ave to h ave comp

etit ive adva ntag e. W itho ut c ompe titi ve a dvan tage ma ny p eopl e wi ll.. . Ot hers can eas ily take you r ma rket awa y. E spec iall y if you hav e mo re r esou rces . Wh at a re t he t hree pri orit ies? The mar ket is b ig, non- comp etit ive, and adv anta geou s. I thi nk... The mark

et i s a matt er o f wh ethe r it 's w orth it or n ot. And then the pro blem of non- cons umpt ion is w heth er y ou s till hav e a chan ce. Timi ng i s ve ry i mpor tant . If you do it o nce it b

ecom es a con sens us, then the re i s ba sica lly no c hanc e fo r yo u. U nles s yo u ha ve r esou rces ten tim es a s mu ch a s ot hers . So you see us uall y it 's t he b ig m arke t th at c an c

atch up when it' s be hind . An d th en t here 's a noth er o ne, whet her you are real ly i nter este d in thi s th ing. Thi s is als o so meth ing to c onsi der. Lik e it or not, thi s is a m atte

r of whe ther you wan t to do it o r not. I th ink you have to choo se b etwe en w heth er y ou w ant to d o it or not, and whe ther you can do it. And ther e ar e fa ctor s to con side r. C hoos

e on e th ing to d o. B ecau se t here are too man y pe ople in this wor ld w ho w ork hard . Th

e pe rson who can win in the end must be the most har dwor king per son. Wha t de cide s wh ethe r yo u wo rk h ard or n ot? Your own phy sica l co ndit ion, of cour se. But this is a ma tter of

luck . Yo ur g enes yo ur g rowt h en viro nmen t d eter mine s wh ethe r yo u ca n wo rk 1 6 ho urs

luck . Yo ur g enes yo ur g rowt h en viro nmen t d eter mine s wh ethe r yo u ca n wo rk 1 6 ho urs

a da y o r 12 hou rs, or 8 hou rs a day. Whic h is fin e. W hich is fine. Righ t? T hen your bus

ines s wi ll b e go od. And your fam ily will be stab le. Righ t? Y ou c an't say tha t I have par ents to take car e of I have chi ldre n to tak e ca re o f a nd I 'm s till sta rtin g a busi ness

. Th at's ver y ha rd t o ba lanc e. A t le ast you can' t do a h igh- leve l bu sine ss. You can

. Th at's ver y ha rd t o ba lanc e. A t le ast you can' t do a h igh- leve l bu sine ss. You can

do a low -lev el b usin ess. If you do t his busi ness in the AI i ndus try, the n yo u ar e re ady to w ork 24 h ours a d ay. You have to be r eady ri ght? You hav e to be in a sta ndby sta te a ll t

he t ime. Aft er t his, I t hink you hav e to be in t he m iddl e of thi nkin g an d th e ma rket.

Find a f ocus . Fi nd a foc us i n th e wo rk y ou w ant to d o an d in the mar ket. Is it i mpor tant whe ther you are goo d at it or n ot? You can lear n. I t's seco ndar y. I 've been thi nkin

g ab out this for a l ong time . I don' t ha ve a goo d an swer . We cho ose to s tart a b usin ess or n ot. Shou ld I for get what I'm goo d at ? I thin k if you do well yo u sh ould n't forg et w

hat you' re g ood at. I'm thin king abo ut t his too. For exa mple if I'm too imp atie nt, shou ld I kee p do ing it? I've bee n th inki ng a bout it late ly. I fe el l ike I ca n ke ep d oing it. Bec

ause it seem s th at t here are sti ll a lot of thin gs t hat Taij i ca n do in LIM. If we d o th is w ell, it will act uall y ha ve m ore valu e fo r so ciet y. F or t he c ompa ny, we c an a lso

use Taij i as our mac hine 's t echn ical fac ilit y. I f yo u sa y th at a per son is g ood at s omet hing or not, I d on't thi nk y ou n eed to b e pa rtic ular ly g ood at i t. L et m e gi ve y

ou a n ex ampl e. F or e xamp le, look at the most suc cess ful CEOs now . Ho w ma ny o f th em a re b orn with a t echn ical tal ent? They mos tly have a v ery stro ng t echn ical bac kgro und. For

exa mple Ji nson Hua ng, he s tudi ed E E. A nd t hen if y ou l ook at C hina li ke L ei J un, he u sed to b e a Chin ese vill age old man, who wro te c ode. Is his code lev el r eall y to p-no tch?

I d on't thi nk s o. I thi nk y ou c an d o it if you' re g ood at i t. I f yo u're not goo d at it at a ll, for exam ple, I u sed to s tudy pro fess iona lism my con nect ions my exp erie nce, and

I'm not a b ig g uy a t all. The succ ess rate is actu ally ver y lo w. F or e xamp le, look at Ma Y un. Ma Y un d oesn 't s eem to b e a tech nica l pe rson bu t he kno ws E ngli sh. He s peak s En

un. Ma Y un d oesn 't s eem to b e a tech nica l pe rson bu t he kno ws E ngli sh. He s peak s En glis h ve ry w ell. Or look at Elon Mus k. H e st arte d to wri te g ames whe n he was ver y yo ung.

He has a te chni cal back grou nd w hen he w as v ery youn g. T he b est CEOs are all fro m te chni cal back grou nds. I t hink my tech nica l ab ilit y is a l ittl e be tter tha n th eirs . I don'

t ne ed s o ma ny t echn ical bac kgro unds . So I w ant to a dd s ome know ledg e on bus ines s. B

ut I thi nk t his is n eces sary. 但是你不 需要成为 这个星球 上最擅长 这个事的 人 才 把它变成 一个bu sine ss。 对,我换 一个说法 说 你 是这个星

球上最擅 长这个事 情的人, 也没有 意义。 因为你作 为CEO 有太 多的事情 必须 是CEO 这个职位 去做。 比如说大 客户, 比如说融 资 比 如说招聘 。 那么 你比如说, Even if

you are the best at writ ing CUDA and PTX wh at c an y ou d o? C an i t he lp y ou r ecru it i nves tors ? If it work s m aybe . Bu t yo u wo n't have tim e to do it. You can' t do it if y ou a re t

tors ? If it work s m aybe . Bu t yo u wo n't have tim e to do it. You can' t do it if y ou a re t he b est at i t. I f yo u we re H u Yu anmi ng i n 20 25, you woul d le ave the inve stor com pany

you men tion ed. What wou ld y ou d o if you wer e a star tup? Tha t's a go od q uest ion. I w ould

exp lore wha t bi g th ings the AI can do i n th e co nsum er s cene. AI f or f un. AI f or f un.

Yeah.

That 's i t. L et m e ad d so me c onte xt. We'v e be en t alki ng a bout thi s re cent ly. Yush

u is pro babl y th e mo st s ucce ssfu l ro bot manu fact urer in Chin a. I t's a hu man- base d ro bot indu stry . It 's r ecei ved more tha n 1 bill ion RMB. Wha t do es t hat mean ? It 's o ne t

enth of the bubb le m arke t. B ut Y ushu has fin ally rea ched the top of the indu stry . It 's d one so w ell. Peo ple all over the wor ld b uy t heir rob ots. But the bub ble mark et i s ob viou

sly doin g be tter. This is toda y r ight ? Ye s t his is t oday . Bu t it may not be like thi s in the fut ure. Not nec essa rily . Th is i s li ke w hat I ju st s aid. Lav able Con cert it qui ckly

hit the AR of 1 00m, 200 m. I s it bec ause we are doin g th is b usin ess now, it' s no t go od w itho ut t hem? I d on't thi nk s o. Y ou a re a lway s on the tab le, you have a g ood team yo u ha

ve a cha nce to d o so meth ing new. You hav e a stea dy b usin ess. In fact, You alwa ys h ave a ch ance to do s omet hing lik e NV IDIA . As lon g as you hav e th is g ene, the gen e is act uall y ve

ry i mpor tant . Is a c ompa ny a tec hnic al g ene or a mar ket gene or a ch anne l ge ne? This is

actu ally ver y di ffer ent. Wha t is the gen e of Ash i? W e ar e a comp any with a t echn ical gen e. D o yo u th ink this gen e is a g ood thin g or a b ad t hing for you as a bu sine ss? What gen

e. D o yo u th ink this gen e is a g ood thin g or a b ad t hing for you as a bu sine ss? What gen

e do you thi nk n eeds to be a dded? I th ink it's a g ood thin g. I thi nk w e ne ed m ore tech nolo gy i n th e en tire tea m. B ecau se I thi nk o ur p rodu ct d esig n m arke t a nd c hann els

are all pret ty g ood. The y ar e ju st m y we ak p oint s. S o n ow, the comp any is q uite cap able . Yo u ju st m enti oned a g ene. I t hink the cul ture of the comp any is t he g ene. What kin

able . Yo u ju st m enti oned a g ene. I t hink the cul ture of the comp any is t he g ene. What kin

d of rul es d o yo u th ink you want to keep in Mesh ? Wh at k ind of c ultu re d o yo u th ink you gave up and chan ged afte r 4 or 5 yea rs? I th ink the most imp orta nt g ene is t he r eal worl

d. Y ou w ill see a lo t of mis take s. T he b ig r easo n is tha t yo u do n't see the real wor

d. Y ou w ill see a lo t of mis take s. T he b ig r easo n is tha t yo u do n't see the real wor ld o bjec tive ly. This has alw ays been the bas is o f al l cu ltur es i n ou r co mpany. Resp ect

fact s. 尊 重事实。 比如说 这事你做 的好,好 就是好, 不好就是 不好,尽 可能客观 去看待它 ,而不是 把不好说 成好,好 说成不好 ,或者说 对外传播 一些没有 数据支撑 的信息,

来夸大自 己有多好 ,我觉得 这个其实 是我们的 文化里面 不会去做 的事情。 犯过这 样的错吗 ? 不尊 重事实在 曼西里面? Yes, it is. It i s re late d to mod el e valu atio n. F or e xamp le, some time s yo u ac cide ntal ly m ake this mis take . Fo r ex

ampl e y ou d on't kno w if thi s th ing is g ood or n ot. You don' t ha ve a n ob ject ive way to e valu ate it. So, in t he e nd, the stud ents who bec ome a tr ee m odel wil l th ink that the ir m

odel is real ly g ood. But in fact yo u ha ve t o ha ve a sys tem eval uati on m etho d to eva luat e wh ethe r it is good or not. Thi s is act ually You have to be a s ob ject ive as p ossi ble. Bec

ause eve ryon e ha s th eir own self -pro tect ion. Thi s ki nd o f se lf-p rote ctio n wi ll m ake ever yone fee l th at t hey are doin g a good job . Bu t if you fin d an othe r pe rson to pres ent

rand omly th is i s a new vers ion of M esh, thi s is an old vers ion, ask him whi ch v ersi on i s be tter . He doe sn't kno w wh ich is n ew a nd w hich is old. Som etim es, you will do it a nd y

ou w ill find tha t th e ol d ve rsio n is bet ter. We have don e so muc h i t ma kes no s ense . Bu

t if you say, I be liev e in som e an cien t my ths that har m pe ople . Fo r ex ampl e o ne p oint of the root cau se, one poin t of the res ult. I t hink thi s is com plet ely dece ivin g. F or e

xamp le, the Heav ens are full of hatr ed. I th ink this is dece ivin g. I f yo u be liev e th at I hav e wo rked so hard I will def init ely get a re ward . Th is i s co mple tely ill ogic al. The

worl d is ful l of eff orts . No one wil l gi ve y ou a rew ard. The mor e yo u do th e gr eate r th e ri sk. Ofte n yo u ha ve w orke d 99 poi nts, but you don 't g et a rew ard. But your min d wi

ll p rote ct y ou. You will say I work ed s o ha rd, how come I d idn' t ge t an y fe edba ck? I sh

ould hav e im prov ed a lit tle. As long as you have thi s ki nd o f th ough t y ou w ill not face the rea lity in the futu re. The bigg est crue l fa ct o f en trep rene ursh ip i s th at m ost of y

our effo rts are usel ess. But you hav e to mak e th is e ffor t. B ecau se y ou d on't kno w wh ich one of y our 100 effo rts is u sefu l. Y ou h ave to t ry t he r emai ning 99 poin ts. But you also

hav e to res pect thi s fa ct. This fac t is cal led, mos t of my atte mpts are use less. In t he M ESHI cha pter I have ano ther que stio n ab out this era . Be caus e o r I shou ld s ay i t th is w

ay, the 3D p rodu ctio n bu sine ss h as a lway s ex iste d. 2 0 ye ars ago, you mad e it an d so meon e bo ught it. 10 year s ag o y ou m ade it, and some one boug ht i t. B ut t oday th e er a is

dif fere nt. The prod ucti on i s AI . We hav e LO M. I n th is e ra, in t he M ESHI bus ines s w hat oppo rtun itie s an d ch alle nges do you have ? Ho w do you pos itio n yo urse lf? Are you worr

ied that som e ne w te chno logy wil l co me o ut o ne d ay a nd y ou w ill be o vert hrow n? Y es,

that 's a ver y go od q uest ion. I th ink ther e ar e ma ny o ppor tuni ties . Of cou rse, no matt er w hat tech nolo gy, many tec hnol ogie s ar e ge ttin g ma ture . Li ke D iffu sion RF DI T o

r MO E te chno logy can mak e it pos sibl e to mak e su ch a mod el. On t he o ther han d t here are man y up stre am a nd d owns trea m in dust ries tha t ar e gr adua lly beco ming mat ure. For exa

mple Te xtIm age, lik e Bl ack Fore st L abs, the y ar e do ing very wel l. F or e xamp le, Lang uage Gen erat ion, lik e Ch achi BT, Clou d t hey are doin g ve ry w ell. and Gemi ni's Goo gle'

s Vu e 3. 0. T here are man y th ings you don 't n eed to b uild . Yo u ju st n eed to f ind a su ppli er a nd g et t he A PI i n. T his is a big opp ortu nity . Th e bi gges t ch alle nge is t hat

the pace of this era is very fas t. S o yo u ha ve t o wo rk v ery hard to buil d an d up date you r pr oduc ts t o ma inta in a bet ter posi tion. I th ink this is the same for all AI comp anie s. H

ow i s yo ur w ork now? My work is quit e st rong . I don' t th ink it's app ropr iate to talk abo ut i t in thi s sh ow. Othe rwis e e very one may not dare to star t a busi ness . Ar e yo u ti red

of P hD o r no w? O f co urse I' m ti red of P hD. PhD was the happ iest tim e fo r me aft er I cha nged my ment or. Beca use I ju st h ave to t ake care of myse lf. But now as a CEO yo ur r espo nsib

ilit y is ver y im port ant. The whol e co mpan y is goi ng t o wi n e very one has to h ave a me al, and the inve stor s wi ll l ook forw ard to t he c ompa ny's rap id d evel opme nt. The cust

omer s ho pe t hat you can deli ver your new pro duct s as soo n as pos sibl e. A fter the com pany gro ws, ther e ar e st ill many thi ngs that I a m no t so goo d at dea ling wit h w hich I h ave

to l earn to deal wit h no w. S o I thin k t o pu t it blu ntly it is a ve ry t irin g st ate. But

eve ry d ay w hen I ge t up I feel tha t th is d ay i s go od e nough. I ca n st ill go a nd t ake on n ew c hall enge s. A nd d urin g th is t ime, eve ry t wo m onth s I fee l li ke I 've impr oved

a l ittl e bi t. I thi nk d urin g th is t ime, I'v e ac tual ly i mpro ved a lo t of my flaw s in my pers onal ity or a lot of my b lind spo ts i n my per cept ion. The re's a p hras e th at I fin

d ve ry i nter esti ng. It's cal led, "Th ere' s no thin g mo re i mpor tant tha n st arti ng a bus ines s th an t rain ing one pers on." Jen sen said he woul dn't sta rt a vid eo a gain . Wh at a

bout you ? I thin k it 's j ust a ma tter of time. I th ink you have a c ompa ny w orth $40 ,000

bout you ? I thin k it 's j ust a ma tter of time. I th ink you have a c ompa ny w orth $40 ,000 . Yo u ca n sa y an ythi ng. You are a fa ther . I will cho ose to w ork in t his comp any agai n. A

. Yo u ca n sa y an ythi ng. You are a fa ther . I will cho ose to w ork in t his comp any agai n. A

ltho ugh I su ffer a l ot i n th is p roce ss, I al ways fee l th at I can lea rn a lot in this pro cess . I thin k I have bec ome a be tter me. I ca n se e th e wo rld in a bro ader per spec tive tha

cess . I thin k I have bec ome a be tter me. I ca n se e th e wo rld in a bro ader per spec tive tha n th e sm all cant een I us ed t o be in the acad emic wor ld. And it s ound s li ke t he l esso

ns y ou l earn ed a re t hing s th at y ou c an't lea rn a s a prof esso r or a s tude nt. I th ink so. You can' t ex pect to lear n to o mu ch i n a prot ecte d en viro nmen t un less you hav e a

so. You can' t ex pect to lear n to o mu ch i n a prot ecte d en viro nmen t un less you hav e a good men tor. Wha t do you hop e to be reme mber ed b y pe ople in five or ten year s? I s it you

r te chno logy yo ur p rodu ct, or y our cult ure? I t hink ... It s houl d be a p rodu ct t hat is t echn olog ical ly d rive n. A com pany 's c ultu re i s al ways to win serv ices . If thi s co

mpan y's cult ure make s th is c ompa ny u nabl e to win th en w e sh ould giv e up suc h a cult ure. But thi s do es n ot m ean that you hav e to ado pt a cer tain typ e of wol f-li ke c ultu re o

ure. But thi s do es n ot m ean that you hav e to ado pt a cer tain typ e of wol f-li ke c ultu re o r so meth ing that eve ryon e ha tes. Ins tead yo u ha ve t o fi nd a min orit y th at a gree s wi

th y ou. a cu ltur e of coh esio n. T his cult ure can be t hat we a ll w ork hard we are res pons ible we do thin gs w ith a ve ry c onsi sten t pu rsui t w e ca n ta ke r isks or emb race cha nges

. Th ese are just pro cess es. A go od c ultu re m ust be u sed by a goo d co mpan y a goo d ta lent

. Th ese are just pro cess es. A go od c ultu re m ust be u sed by a goo d co mpan y a goo d ta lent tea m a goo d pr oduc t. B ut a s a CEO, you mus t ch oose a c ultu re t hat can win. If y ou c hoos

e a book cul ture yo u do n't need to use it. What are the cha ract eris tics of a bo ok c ultu re?

Book cul ture in trov erte d i mpra ctic al. I sa y im prac tica l n ot b ecau se w e ha ve h igh goal s. I thi nk t he g oal can stil l be hig h. B ut t here is one thin g ab out book cul ture it

goal s. I thi nk t he g oal can stil l be hig h. B ut t here is one thin g ab out book cul ture it doe sn't fac e th e fa cts. I d o th is, I mu st m ake it i nto some thin g el se. Or i f so meth

ing goes wro ng, I do n't go b ack to i t I don 't r efle ct o n it. Or t he t eam can' t cr itic ize the CEO, can 't p ubli cly poin t ou t ot her peop le's pro blem s. E very one seek s ha rmon

y. I thi nk t hese are all the way s to los e. D o yo u of ten get crit iciz ed b y yo ur t eam? I w

y. I thi nk t hese are all the way s to los e. D o yo u of ten get crit iciz ed b y yo ur t eam? I w

as j ust scol ded by s omeo ne t he o ther day . Wa s it one -on- one or i n pu blic ? Ab out last wee k t his clas smat e sa t he re a nd t he o ther cla ssma te, thre e pe ople pu lled me to t he c

orri dor and said I h ad s omet hing I d idn' t do wel l. I t's not in p ublic. It's lik e a one- on-o ne o r so meth ing. Wha t ki nd o f th ings ? Ba sica lly, bus ines s-wi se, team man agem ent-

wise ? Ma nage ment -wis e. Y ou'r e ve ry a ccep ting of this kin d of thi ng. Yes, I h eard it

wise ? Ma nage ment -wis e. Y ou'r e ve ry a ccep ting of this kin d of thi ng. Yes, I h eard it at f irst . I was very ale rt, but I wo uld list en t o hi m. T hen I th ough t o h a m I real ly n

ot d oing wel l? L ater I f ound out tha t th ere were ind eed plac es w here I w asn' t do ing well . I thin k yo u're ver y lu cky to h ave such an atmo sphe re t o me et s uch peop le. Yes, yes ye

. I thin k yo u're ver y lu cky to h ave such an atmo sphe re t o me et s uch peop le. Yes, yes ye s. B ecau se m any peop le d on't spe ak up. He m ust have a s tron g ow ners hip of t his comp any.

s. B ecau se m any peop le d on't spe ak up. He m ust have a s tron g ow ners hip of t his comp any.

And he hope s th is c an b e do ne. He w ill be w illi ng t o te ll y ou t his when he sees him self as the CEO. Yes I thin k th is i s a basi c wo rk. If y ou s ay i t's a ma nufa ctur ing comp any,

the n yo u ca n be mor e po werf ul. But we a re n ot. We a re a com pany tha t dr ives inn ovat ion. You have to acce pt t hat the CEO does n't know eve ryth ing abou t th e co mpan y. I wan t to

ion. You have to acce pt t hat the CEO does n't know eve ryth ing abou t th e co mpan y. I wan t to tal k ab out a to pic call ed " Goal s" a bout Mas h. L ast year Li Mu gave a t alk. He said tha

t th ere are diff eren t st ages of goal s. H e sa id t hat stud ents sho uld stud y ha rd a nd p ass GPA and pass pap er. KPI is t o pa ss p aper . Wo rkin g is pro moti on. He s aid that the goa

l of ent repr eneu rshi p is exi t. I s ex it y our goal of entr epre neur ship? I th ink it's a m eans no t a goal . Of cou rse, I w ill say that aft er 3 or 4 ye ars, we have to IPO this as our

goal . It see ms t hat it c an b e ac hiev ed n ow, if i t go es w ell. You hav e to ans wer afte

goal . It see ms t hat it c an b e ac hiev ed n ow, if i t go es w ell. You hav e to ans wer afte r th e IP O. O r if you can 't I PO, then you won 't d o it . I thin k o f co urse so meon e ha s me

ntio ned the acqu isit ion or s ome othe r th ings bu t we are sti ll h ere, sti ll c onti nuin g to do it o urse lves.

For me, maki ng a pro duct tha t ev eryo ne l oves or in the proc ess, con stan tly gett ing to k now myse lf m ore, and bec omin g a bett er p erso n a nd l eadi ng e very one in t he t eam to b

ecom e a bett er p erso n i s th e ve ry i mpor tant pur pose of this com pany . As for wha t it wil l be lik e in the end I thin k it 's e noug h to let eve ryon e t o be abl e to fac e ev eryo ne.

Ever yone wil l ha ve a goo d fi nanc ial retu rn, and a ca reer dev elop ment . We als o ma de a pro duct tha t we are pro ud o f. I think make Mas sey a gr eat plac e to wor k. M ake Mass ey a gre

at p lace to work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas sey a gr eat plac e to wor k. M ake Mass ey a gre at p lace to work . Ma ke M asse y a

rk. Make Mas sey a gr eat plac e to wor k. M ake Mass ey a gre at p lace to work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas sey a gr eat plac e to wor

k. M ake Mass ey a gre at p lace to work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey

k. M ake Mass ey a gre at p lace to work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas sey a gr eat plac e to wor k. M ake Mass ey a gre at p lace to

work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas

work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas

sey a gr eat plac e to wor k. M ake Mass ey a gre at p lace to work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas sey a gr eat plac e to wor k. M ake Mass

ey a gre at p lace to work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas sey a gr eat plac e to wor k. M ake Mass ey a gre at p lace to work . Ma ke M

asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas sey a gr eat plac e to wor k. M ake Mass ey a gre at p lace to work . Ma ke M asse y a grea t pl ace to w ork. Mak

e Ma ssey a g reat pla ce t o wo rk. Make Mas sey a gr eat plac e to wor k. M ake Mass ey a gre at p lace to work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk.

Make Mas sey a gr eat plac e to wor k. M ake Mass ey a gre at p lace to work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas sey a gr eat plac e to wor k. M

ake Mass ey a gre at p lace to work . Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas sey a gr eat plac e to wor k. M ake Mass ey a gre at p lace to work

. Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas sey

. Ma ke M asse y a grea t pl ace to w ork. Mak e Ma ssey a g reat pla ce t o wo rk. Make Mas sey a gr eat plac e to wor k. M ake Massey They hav e a lead er w ho c an l iste n to the ir d eman ds.

They pla yed a mo re i mpor tant rol e in the dev elop ment of thei r co mpan y. T hey also kno w th e mo st a dvan ced tech nolo gy. Mayb e in the pas t t hey only wro te a sha der in t he f acto ry,

and they cou ld p lay with 512 GPU s. E very one has beco me a bet ter pers on d urin g th is p roce ss. Yuan ming I' ve a sked you abo ut t he M esh part . I want to give you som e ti me t o ad d an

ss. Yuan ming I' ve a sked you abo ut t he M esh part . I want to give you som e ti me t o ad d an ythi ng y ou w ant to t alk abou t Me sh o r ev en a nyth ing you want to recr uit in t his podc

ast. Oh, tha t's grea t. W e've bee n re crui ting AI rese arch ers, mac hine lea rnin g en gine

ast. Oh, tha t's grea t. W e've bee n re crui ting AI rese arch ers, mac hine lea rnin g en gine ers, and com pile r en gine ers. Our mai n re crui tmen t is sti ll i n th e AI par t of Mes h. A

nd i n th is p art, we have a v ery gene ral pack age. and our inte rns are also Glo bal Pay, the hig hest sal ary is 1 mil lion RMB per yea r or 140 ,000 USD . I will spe nd t ime with the int erns

to do t he p rodu ct a nd t he r esea rch. If you are stil l a stud ent, you are wel come to appl y fo r ou r in tern ship pro gram . Of cou rse, our ful l-ti me o ffer wil l be mor e ge nero us t

han this . Wh ethe r yo u li ke T rain Mod el or We h ave a lo t of pos itio ns h ere. Wel come to

han this . Wh ethe r yo u li ke T rain Mod el or We h ave a lo t of pos itio ns h ere. Wel come to sign up. No matt er w here you are in the worl d w e ar e a glob al d istr ibut ed c ompa ny. I am

sign up. No matt er w here you are in the worl d w e ar e a glob al d istr ibut ed c ompa ny. I am

in Guig u b ut t here are man y pe ople in othe r pa rts of t he w orld . We hav e be en w orki ng o n th e Ta iji proj ect for a lo ng t ime. to r esto re t his proj ect. Bec ause the cor e of thi

s pr ojec t ha s al ways bee n th ere. The re i s a plac e in Mes h to use TSM C. W e al so w ant to r ecru it t wo o r th ree stud ents who are int eres ted in G PU, prog ramm ing or g raph ics

engi neer ing to w ork with me to m ake TSMC 2.0 a s ucce ss. I th ink with TSM C's curr ent user bas e w e ca n st ill make TSM C a grea t in flue nce in t he f utur e in the fie ld o f LA M an

d AI. One inte rest ing data is that if you look at the numb er o f st ars in t he I nvid ia W arp,

d AI. One inte rest ing data is that if you look at the numb er o f st ars in t he I nvid ia W arp, Tri ton, and Mod ulo proj ects th ere are not many Tai ji. Taij i is sti ll v ery popu lar, but it has not foun d a big fiel d to dem onst rate . So I t hink in the 2.0 era, Tai ji w ill defi nite

ly m ake a bi gger imp act. Glo bal recr uitm ent. And you r co mpan y sh ould be very fle xibl e. W

ork from hom e sh ould be... Okay. 其实前三 个词是来 自于老黄 的。 老 黄有一次 在Sta nfor d演讲, Stud ents ask ed h im w

hat the stan dard s we re w hen he r ecru ited peo ple at N VIDI A. H e sa id, "Bri ng G od a nd H eart ." " Brin g" i s ac tual ly e asy to u nder stan d. W e al l ho pe t hat the stro nges t pr

eart ." " Brin g" i s ac tual ly e asy to u nder stan d. W e al l ho pe t hat the stro nges t pr oble m so lver can joi n ou r te am. "God " me ans that you hav e to tak e ac tion qui ckly whe n th

e in form atio n is inc ompl ete and prom ote some thin g. Y ou c an a ccep t th e ri sk o f fa ilur e an d le arn from fai lure . "H eart " me ans that we hope you are kin d fr om t he b otto m of you

r he art. Tast e. T aste. Then wha t bo oks to r ead? It' s ha rd t o de scri be i n wo rds, but onl

y pe ople wit h th e sa me o rder can wor k cl osel y to geth er f or a lon g ti me. What do you do o n we eken ds? What boo ks d o yo u re ad? I've bee n re adin g mo re b ooks by Xiao mi L ei J

un r ecen tly. I t hink he' s a very wor th l earn ing entr epre neur . He 's a ctua lly. .. I thi

nk t here are man y as pect s th at I 'm a bit lik e hi m. H e sa id h e wa s a vill age old devi l. W hen he w as b usy play ing with Mic roso ft, he w as d oing Int erne t. T hen he c ompl etel

l. W hen he w as b usy play ing with Mic roso ft, he w as d oing Int erne t. T hen he c ompl etel y mi ssed the Int ernet. and fina lly real ized a t heor y of fly ing pigs . Th e pi g ca n fl y up

the typ hoon . Ac tual ly, the "why " I ment ione d is sim ilar . He was ver y di lige nt. At t hat time he did the off ice with Mic roso ft. He m ade a se t of sto cks and fina lly didn 't s ell them

. It als o re mind ed m e of the ear ly d ays of e ntre pren eurs hip, whe n I was maki ng p rodu

. It als o re mind ed m e of the ear ly d ays of e ntre pren eurs hip, whe n I was maki ng p rodu cts that fai led to b ecom e co mmer cial ized . Bu t he is now a ve ry s ucce ssfu l en trep rene

ur. I th ink his purs uit of p rodu ct q uality for the cont rol of d etai ls, for the cont rol of d

ur. I th ink his purs uit of p rodu ct q uality for the cont rol of d etai ls, for the cont rol of d irec tion an d fo r th e re cove ry o f th e pr evio us s etba cks. The y ar e al l ve ry h elpf ul.

You ment ione d th at y ou d idn' t go to FAIR . Bu t I want to dig deep er o n th is p oint . Be

caus e yo u sa id t hat "exp lore " is mor e th an " expl oit" . No w h ow d o yo u ba lanc e sh ort- term imp act and long -ter m vi sion ? I thin k th is h as a lway s be en a mat ter of b alan ce.

My g ener al p ract ice is t o sp end 20% of m y ti me t hink ing abou t th e fu ture . Vi sion is a co ntin uous pro cess . It 's n ot l ike I ha ve a vis ion toda y an d it 's o ver. Bec ause the

who le m arke t wi ll c hang e co nsta ntly . Yo ur p erce ptio n wi ll c hang e co nsta ntly . So

may be a bet ter word is visi onin g. Y ou h ave to t hink abo ut w hat the futu re w ill be l ike

ever y mo ment. That 's y our choi ce. They 're will ing to c hoos e. T hat' s it. I ca n on ly t

ell you that you hav e to thi nk a bout it. You hav e to thi nk a bout it. Thi s in dust ry i s mo ving ver y fa st. It m ay b e a litt le h ard to w ork. If you are will ing to a ccep t it yo u wi

ll c ome. If you are not will ing to a ccep t it yo u wi ll n ot c ome. I t hink in the futu re o f th e wo rld, mos t pe ople wil l no t ne ed t o wo rk. What wil l th ese peop le d o? E very one

has the cond itio ns t o fl y a priv ate plan e to exp erie nce what the Sou th P ole is l ike in a ll c orne rs o f th e wo rld. Wha t is the cap ital of Egypt? and even go back to the past to see

what the soc iety of the anci ent Baby lon was like ? I don' t th ink so. So i f we can use the tec hnol ogy of t he m oder n de ep-s eate d AI to crea te a n im mers ive expe rien ce f or e very one,

so that eve ryon e's life ev en i f th ey h ave live d th e sa me l ife, wit h th is t echn olog y y ou c an l ive a ri cher lif e. Y ou c an g o ba ck t o yo ur p ast and do t hing s th at y ou d idn'

t ha ve t he c hanc e to do at t he t ime. May be g o ba ck to I th ink ulti mate ly, this is what

the wor ld n eeds. I wa nt t o ta lk t o yo u ab out a ke y wo rd t hat you ment ione d ca lled sta ge 1 and sta ge 2 . Ye s. Y ou s aid stud ents . Wh y do n't you intr oduc e th is c once pt? Ther

e ar e ma ny e xamp les. Sta ge 1 and sta ge 2 ba sica lly stag e 1 is s olvi ng t he p robl em o f nu mber . St age 2 is sol ving the pro blem of Y. A t st age 1 y ou h ave a cl ear goal . At sta

ge 2 yo u do n't have a c lear goa l. A t st age 1 y ou c an d o yo ur o wn t hing . At sta ge 2 yo u ha ve t o th ink abou t wh at t he w hole tea m sh ould do. At s tage 1, some one else wil l gi

ve y ou p robl ems. At stag e 2, you wil l gi ve y ours elf prob lems . Fo r ex ampl e f rom a un derg radu ate stud ent, the mos t im port ant thin g is to lear n th e kn owle dge in t he t extb

ook. To a gr adua te s tude nt, you may need to find a g ood rese arch top ic. And then th ere are

ook. To a gr adua te s tude nt, you may need to find a g ood rese arch top ic. And then th ere are stag es 1 and 2 i n gr adua te s tude nts. At the earl iest sta ge, the inst ruct or w ill give you

a t opic to stud y. B ut a t th e se cond sta ge, you need to find a g ood topi c yo urse lf. Beca

use the inst ruct or w ill not nece ssar ily have you r fi rst- line per cept ion of t he p robl em.

He m ay n ot k now what the mos t im port ant prob lem is. You can only rel y on you rsel f. F or e xamp le, you used to be a n in divi dual con trib utor in a bi g co mpan y. Y ou j ust need to do t

he t asks tha t th e pr evio us g ener atio n as sign ed y ou. Now you have bec ome a ma nage r. Y

ou n eed to g ive the enti re o rgan izat ion a vi sion gi ve t hem a go al, and let the enti re o rgan izat ion oper ate. You nee d to fac e hu ge u ncer tain ty. For exam ple, you use d to be a co

mpan y em ploy ee, and now you' re s tart ing your own bus ines s. W hen you used to work, All the plat form s an d th e ce rtai nty are prov ided by the comp any. But aft er y ou s tart you r ow n bu sine ss, you lose all of that . Al l th e th ings you do beco me s omet hing you hav e to fig

ure out. The thi ngs you do n ext depe nd o n yo ur o wn e xplo rati on a nd n ew k nowl edge . Th

ure out. The thi ngs you do n ext depe nd o n yo ur o wn e xplo rati on a nd n ew k nowl edge . Th

ese two stag es a re s o di ffer ent that man y pe ople wil l co ntin ue t o us e th e fi rst stag e of thi nkin g in the sec ond stag e. T his will lea d to the dis aste r of the sec ond stage. And

many exa mple s ar e pe ople who are ver y su cces sful in stag e 1 and then the y re ach stag e 2.

Yes th e mo re s ucce ssfu l yo u ar e in sta ge 1 th e mo re s uffe ring you wil l ex peri ence in stag e 2. Why ? Be caus e yo ur s kill s in sta ge 1 and sta ge 2 are com plet ely diff eren t. S

tage 1 m ay b e to obe y th e co mman d r oll effi cien cy, and keep on grin ding a s peci fic skil l. B ut i n st age 2 n o on e is com mand ing you anym ore. Eff icie ncy is a lso impo ssib le t

l. B ut i n st age 2 n o on e is com mand ing you anym ore. Eff icie ncy is a lso impo ssib le t o ta lk a bout bec ause the met ric of e ffic ienc y is not def ined . Ef fici ency can not be d

efin ed. You won' t ha ve t he e nerg y to pra ctic e yo ur s peci fic code wri ting dr awin g o r a

efin ed. You won' t ha ve t he e nerg y to pra ctic e yo ur s peci fic code wri ting dr awin g o r a spec ific ski ll. So y ou w ill be v ery, ver y un accu stom ed. If y ou d on't rea lize tha t o

kay, I'm in the seco nd s tage now . It 's l ike play ing a bo ss. One bloo dlin e is fin ishe d n ow i t's goin g to be defo rmed pl ay t he s econ d bl oodl ine. You sti ll u se t he f irst blo

odli ne t o fi ght it, it w ill be o ver soon it wil l co me b ack. Hav e yo u ma de s uch a mi stak e? I have. I th ink I ma ke t his mist ake ever y mi nute of my l ife. The re a re s tage s in lif

e? I have. I th ink I ma ke t his mist ake ever y mi nute of my l ife. The re a re s tage s in lif e th at c hang e ev ery minu te o f my lif e. J ust like I s aid, why do I ha ve t o be come a n

ew m e ev ery thre e to six mon ths? If you don' t be come a n ew y ou e very thr ee t o si x mo nths yo u wi ll a lway s be stu ck i n th e pr evio us s tage . If you r co mpan y in crea ses by 2

0% e very mon th, you want to doub le y our reve nue in t hree or four mon ths. Are you sti ll d eali ng w ith the curr ent prob lem in t he w ay y ou u sed to? No. How do y ou j udge a p erso n? W

ill they bec ome a su cces sful sec ond stag e im pact mak er? It d epen ds o n wh ethe r th ey c an a dmit tha t th ey a re a foo l. R espe ct t he f acts . Re spec t th e fa cts. If they mak e a

mist ake, the y ad mit it. And then the y ke ep i mpro ving the msel ves. Hav e a good rol e mo del.

Eve n if thi s ro le m odel is far away fro m th emse lves . Wh at i s yo ur r ole mode l? A lot

. El on M usk, I t hink . An d th en i n th e ac adem ic w orld my men tor, Fre do. Then in the Chin ese indu stry I thin k Le i Ju n. O r he re, Jin Sen Huan g H uang Ren jun. In the game ind ustr y G

ou M enma o. O r an othe r pe rson in the worl d of Nin tend o H eng Jing Jun ping . He mad e Ga me B oy. I th ink they hav e al l re ache d th e to p in a c erta in f ield . An d th ey h ave thei

r ow n re ason s fo r re achi ng t he t op. I th ink if y ou l ook at w hy h e su ccee ded, to make you rsel f a bett er p erso n e ven if i t me ans that you wer e a fool in the past th at d oesn

't m atte r. S tart ing from now yo u ar e a new pers on. The past you to day' s yo u t omor row' s yo u h ave noth ing to d o wi th it. you can beco me a bet ter vers ion of y ours elf. Wha t ar

e yo ur s ugge stio ns f or P hD s tude nts in s tage 1 a nd s tage 2? For PhD stud ents in stag e 1, jus t se nd o ut y our pape r. I f yo u do n't have a p aper do n't talk abo ut w hat the tren

d is in this ind ustr y. S tage 1 i s to rol l. R oll and exec ute. I w ill exec ute this wel l. S

end out a fe w mo re p aper s. G et m ore feed back and mor e kn owle dge. And then whe n yo u ha ve t hree or four top pap ers, you hav e to sta rt t hink ing abou t th ese pape rs. Alth ough the

y ar e al l th e sa me, it's a p aper ri ght? How did I d o it ? Ho w ma ny p eopl e ha ve u sed my w ork? Whi ch i ndus trie s ha ve b een chan ged? Are the se j obs in t he s yste m? I s it a j

ob t hat ever yone rec ogni zes as t he b est job? How man y pe ople hav e ad opte d th is? For exam

ple, Res Net is a plu s. H ow m any peop le h ave adop ted this plus? And then ar e th ere any othe r pe ople in the indu stry who use our wor k? I s th ere a ga p? O nly with the cra zy r oll of s

tage 1, you can get enou gh i nfor mati on t o br ing you to s tage 2. But afte r st age 2 e xecu tion is stil l ve ry i mpor tant . Bu t it tak es t ime to t hink abo ut w hy. Many tim es, ever

yone is stil l dr inki ng w ater at stag e 2. I h ave thre e fi lms now, I w ill post thr ee f ilms thi s ye ar, and it w ill be s ix f ilms . An d th en I kee p do ing this . I can do i t in man y wa

ys, but just lik e wh at y ou j ust said, Many thi ngs with 10 time s th e im pact are onl y 2 time s ha rder to do. What are the mor e su cces sful PhD s yo u've see n? T hat' s a good que stio n. I

thi nk i t's defi nite ly K evin . Of cou rse, he' s al read y a prof esso r. Y ou o nce wrot e th at y ou s aid not only do the smal l fi sh i n th e se a gr aph, but als o th e bi g fi sh i n th

e en tire CES pon d. W hat is t he i nflu ence you def ine now? Wha t is the big fis h? H ow d o yo u ju dge whet her you have mad e a big fish ? Th at's a g ood ques tion . I thin k if Tai ji 2 .0 c

an b e us ed i n th e er a of the big fis h an d th e ne w mo del, the re a re s till man y ap plic atio ns, and its infl uenc e is Tai ji 1 .0's 50 t imes mor e th an t hat, I t hink it' s a big fish

. Wh at w ill the 2.0 Taij i lo ok l ike? May be a ll t he t rans form ers will use som ethi ng l

. Wh at w ill the 2.0 Taij i lo ok l ike? May be a ll t he t rans form ers will use som ethi ng l ike this . Or we can comb ine the simu lati on a nd t he n eura l ne twor k. W e ca n ma ke a hyb

ike this . Or we can comb ine the simu lati on a nd t he n eura l ne twor k. W e ca n ma ke a hyb rid simu lato r. I t ha s bo th s imul atio n an d da ta-d rive n pa rts. But you hav e to com bine

the se t wo t hing s. T here mus t be goo d in fras truc ture . Ta iji can beco me t hat infr astr

uctu re. In t he r obot ics field, I th ink you' re r ight.

uctu re. In t he r obot ics field, I th ink you' re r ight.

I th ink I wi ll n ever fai l. A nd I don 't t hink I w ill comm it s uici de e ven if I fai l in my comp any. As long as I ca n li ve, I ca n ke ep d oing new thi ngs. So any fail ure is j ust a te

comp any. As long as I ca n li ve, I ca n ke ep d oing new thi ngs. So any fail ure is j ust a te mpor ary fail ure. The rea l fa ilur e is may be t ryin g on ce o r tw ice and then not wil ling to

inno vate aga in. One move was bit ten by a sna ke, ten year s af raid of bein g al ert, and alw ays doin g so me c onse rvat ive thin gs. I th ink it's a r eal fail ure. If your goa l is to inno vate

th en s ince mos t of the inn ovat ions don 't m ake sens e i f yo u ar e su re t o su ccee d t hen don' t in nova te. It's goo d to be amon g te n in nova tion s. I oft en c omfo rt m y te am m embe

rs, sayi ng, it's oka y to fai l in thi s ti me. As l ong as y ou w ork hard an d yo ur i nsig ht i s ri ght, the n fa ilur e is nor mal, plu s th e fo od. It's too eas y to fail. You have a s ente

nce, act uall y y ou s aid that you thi nk l ife is a big dea l a nd e very big thi ng h as t o be don e fo r fo ur o r fi ve y ears . Do you thi nk f rom this per spec tive yo u ha ve d one a fe

w bi g th ings so far, and you sti ll w ant to d o th ose big thin gs i n th e fu ture ? We can fir st c alcu late the num ber. I s aid that bec ause I a ssum ed t hat I st arte d do ing thin gs w hen

I wa s 20 . I grad uate d fr om c olle ge a t th e ag e of 20. I s tart ed d oing thi ngs and reti red at t he a ge o f 70 . Ho pefu lly, I c an r etir e at the age of 70. In t he m iddl e i t's 50 y

ears. In 5 0 ye ars, if you do s omet hing for 5 y ears yo u ca n do 10 thin gs. And then the re m

ears. In 5 0 ye ars, if you do s omet hing for 5 y ears yo u ca n do 10 thin gs. And then the re m ust be h alf of t hese 10 thin gs t hat are fail ed. Beca use if y ou d o al l 10 thi ngs, tha t me

ans you are doin g ve ry l ittl e. T hat' s no t a real inn ovat ion. A r eal inno vati on w ill defi nite ly h ave a fa ilur e ra te. So t hese 10 thin gs m ay o nly beco me 5 thi ngs. How man

y of the se 5 thi ngs have I d one so f ar? You are this que stio n. H ow m uch have you don e no w? W hat do y ou w ant to d o in the fut ure? I t hink I h ave done 0.5 plu s 0. 5 now. One 0.5 is T

w? W hat do y ou w ant to d o in the fut ure? I t hink I h ave done 0.5 plu s 0. 5 now. One 0.5 is T aiji has rea ched thi s st age. I'm pla nnin g wh at T aiji 's n ext 0.5 will be like . If we do t

his well it wil l pr obab ly t ake anot her 4 or 5 y ears to reac h a 1. I thi nk M esh has also rea ched 0.5 . Af ter Perm anen t Ma rket Fit it 's p roba bly from 1 t o 10 fr om 1 0 to 100 . Th ere

ched 0.5 . Af ter Perm anen t Ma rket Fit it 's p roba bly from 1 t o 10 fr om 1 0 to 100 . Th ere

are stil l a lot of t hing s to do. I t hink Mes h's rise to the top is a noth er 0 .5. Afte r th e ac cide nt? Afte r th e ac cide nt, I ca n go bac k to bei ng m y ne xt " one" . Yo u me ntio ned

the abil ity to s olve pro blem s in you r ar ticl e. D o yo u th ink your lif e's mean ing is l imit ed b y yo ur a bili ty t o so lve prob lems ? Ar e th ere any prob lems tha t yo u ar e no t ab

le t o so lve? You rea lize d it and you dec ided not to solv e it . Th ere are many . Fi rst of a ll, I th ink life has man y m any mean ings . It 's n ot t hat you only hav e a very int ense lif

e. I t's mean ingf ul t o li ve a ver y ti ring ha rd-w orki ng, life -cha ngin g li fe. If y our

e. I t's mean ingf ul t o li ve a ver y ti ring ha rd-w orki ng, life -cha ngin g li fe. If y our life 's p urpo se i s to tak e ca re o f yo ur p aren ts, to r aise a w onde rful chi ld, or t o do

vol unte er w ork, to take car e of 100 orp hans th at's als o ve ry m eani ngfu l. I t's not like eve ryon e ha s to go to E lon or J ense n to hav e a mean ingf ul l ife. Eve ryon e ha s th eir own

free dom to c hoos e th eir own life . Th ere' s no suc h th ing as a hig h or low . Es peci ally whe n th e AI era com es, ever yone can eat wel l an d dr ess warm ly, and even don 't h ave to w

ork. I th ink we n eed more tol eran ce i n th is k ind of s ocie ty. More cho ice. Mor e po wer

ork. I th ink we n eed more tol eran ce i n th is k ind of s ocie ty. More cho ice. Mor e po wer to c hoos e. I nclu ding tho se w ho a re l ying fla t no w. I thi nk t hey are lyin g fl at a gain

. Wh at c an y ou d o? Y ou c an l ive a go od l ife. Liv e ha ppil y. T hat' s go od. So, are ther

. Wh at c an y ou d o? Y ou c an l ive a go od l ife. Liv e ha ppil y. T hat' s go od. So, are ther e so me c apab ilit ies that my life cho ices are lim ited by solv ing prob lems ? Of cou rse. I c

an't do XAI like Elo n no w. I don 't h ave the fina ncin g ca pabi lity . I don' t ha ve t he c all powe r. I don 't h ave the mone y to dea l wi th t his. But I'm sti ll h appy to see what my life

powe r. I don 't h ave the mone y to dea l wi th t his. But I'm sti ll h appy to see what my life dev elop ment is like. Mayb e Me sh c an d o a good job and the n go pub lic. The n th ere are more

res ourc es t o do a b igge r th ing. The n yo u ha ve a pub lic comp any and a sm all- scal e fi rst- hand com pany lik e us . Yo u do thi ngs comp lete ly d iffe rent . An d th en I obv ious ly d

on't hav e a mana geme nt e xper ienc e of mor e th an 1 00 p eopl e or 1,0 00 p eopl e. T hen I'm

goin g to gra dual ly l earn the se e xper ienc es. But agai n I thi nk l ikin g so meon e an d be

ing loya l to the m ar e tw o di ffer ent thin gs. A pe rson who liv es t he l ife he l oves wil

l be hap py b ut h e wi ll l ose a lo t of opp ortu niti es. So I thi nk l ife is o kay now. I a m a pers on w ho i s ve ry p atie nt a nd v ery adap tabl e. M y qu alit y of lif e is ver y lo w. I don

't w ant to s tarv e to dea th. If I can do some thin g I' m in tere sted in with out star ving to deat h I thi nk i t's good . So mon ey i s no t yo ur s tart ing poin t? I thi nk i t's only 5% or 1

0%. If I wan t to mak e mo ney, I h ave many way s to mak e mo ney. But aft er m akin g mo ney, I w

0%. If I wan t to mak e mo ney, I h ave many way s to mak e mo ney. But aft er m akin g mo ney, I w ill buy a ho use in A ther ton. The n yo u st ill have to solv e th e pr oble m of wha t I do. It's

mor e im port ant to d o th ings . Ar e yo u st ill writ ing code now ? I do. How much tim e do you spe nd e very day ? I prob ably wro te a few hou rs l ast week . Tw o or thr ee h ours . Th is w

eek, may be a lit tle less . Bu t th is w eek, I h elpe d tw o of my clas smat es h ere debu g. I

fix ed a bug. Do y ou w orry tha t yo u mi ght not be a ble to w rite it one day? I wi ll. I'm worr

ied that I w ill be c ompl etel y on the sam e pa ge. The deci sion s I make are all the sam e as the abo ve. I st ill beli eve that a g ood foun der must be hand s-on . Lo ok a t Le i Ju n J inso

n E lon. The y ar e al l ve ry h ands -on. The y do n't nece ssar ily writ e co de. Yes, the y do n't nece ssar ily writ e co de. They can des ign prod ucts . Th ey c an p erso nall y ta lk t o su ppli ers. The y ca n go to

ppli ers. The y ca n go to I th ink hand s-on is a ba d wo rd. It's abo ut u nder stan ding tec hnol ogy. Lis a Su th e CE O of AMD th e bu ildi ng n ext to u s s he s aid she was at M IT o nce beca use she grad uate d fr om M

IT. She said she was mor e im pres sed by t he f act that she act uall y un ders tand s ho w te chno

IT. She said she was mor e im pres sed by t he f act that she act uall y un ders tand s ho w te chno logy wor ks. She actu ally und erst ands how tec hnol ogy work s. T his is v ery help ful for many

bus ines s ch oices. To w hat exte nt d o yo u ac tual ly u nder stan d? I thi nk I 've done it at l east onc e. F or e xamp le, if y ou d o Me sh, you' ve t rain ed m odel s t rain ed G ANs, tra ined

CNN s e tc. You may not have tra ined Dif fusi on, but you prob ably kno w th e tr aini ng d ynam ics, ADE M O ptim izer an d wh at's the dea l wi th A lpha and Bet a. Y ou w rite Com pile r y ou n

eed to l ead a Co mpil er t eam, you 've at l east wri tten Com pile r be fore . If not wh y wo uld othe rs w ork for you? He c an't lea rn a nyth ing from you . Wh at w ould he do i f Hu Yan ming sto

pped sta rtin g a busi ness one day ? Wr ite a bo ok. Cont inue to make a h igh- end rema ke o f my Gam es 2 01. I wa s li ke y ou g uys. I w as s tudy ing at I nuit a. I rem embe r I was stud ying at

Bost on b ecau se o f CO VID. I r emem ber I st arte d ta lkin g at 8 i n th e mo rnin g. I t wa s 8 in t he e veni ng i n Ch ina. I t alke d at 8 a nd f inis hed at 9 . I slep t un til 12. Then I s

tart ed w orki ng a t 9 a.m. on the West Coa st. I di dn't hav e ti me t o re st o n we eken ds.

I wa s ve ry, very has ty. At t hat time th ere was no s imul atio n in Chi na. So e very one thou ght it w as o kay. But I t hink it was just fro m 0 to 1 . Th ere are a lo t of tec hnic al d etai

ls t hat are not clea r en ough . Be caus e I have lim ited tim e to pre pare . So I s till wan t to do this bet ter. I wa nt t o wr ite a bo ok t hat teac hes ever yone how to do s imul atio n h ow t

o do gra phic s re nder ing. Or the hybr id s imul ator we just tal ked abou t. H ow c an s imul atio n an d ne ural net work be inte grat ed t oget her? And it' s be tter to have a l ittl e ro

boti cs-b ased app lica tion . I thin k th at w ould be bett er. Our podc ast actu ally has a l ot o f yo ung list ener s. T he y oung er g ener atio n. I wan t to kno w... I'm als o ve ry y oung . I

was born in 1994 . I' m no t mu ch o lder tha n yo u. W hat do y ou s ugge st t o yo ung peop le w ho a re g oing to stud y fo r a PhD? Fig ht f or t he t hing s yo u re ally lov e in thi s pr ocess.

I've see n to o ma ny p eopl e re ad b ooks jus t fo r a titl e b ut t hey don' t li ke i t in the end . It 's v ery pain ful. But you act uall y fo und some thin g yo u lo ve, not thro ugh read ing

end . It 's v ery pain ful. But you act uall y fo und some thin g yo u lo ve, not thro ugh read ing book s. Y es, befo re I rea d bo oks, I k new I wo uld like to do t his when I r ead book s. B ecau

book s. Y es, befo re I rea d bo oks, I k new I wo uld like to do t his when I r ead book s. B ecau

se w hen I wa s in the und ergr adua te, I ha d al read y do ne d ozen s of C-g raph pap ers. So

I kn ew I was jus t ma king a p aper fro m th e pa pers of othe rs. I kn ew I wou ld l ike this ver

y mu ch. What do you thin k of you ng p eopl e wh o wa nt t o st art a bu sine ss? Take act ion.

Do y ou t hink thi s ac tion ref ers to e xecu tion or the real ste p to sta rt a bus ines s? I thi

nk t he r eal step to star t a busi ness is that man y th ings can be lear ned. You can qui ckly try thi ngs out, lea rn, and then fin d yo urse lf s tupi d an d be come a b ette r se lf. So e xper

ienc e is not tha t im port ant. It' s no t th at i mpor tant to figu re i t ou t. A s lo ng a s yo u ca n be ar t he c onse quen ces of f ailu re, as l ong as y our ment alit y is tha t fa ilur e is

fai lure wh at c an y ou do? I wa s li ke, okay I grad uate d fr om P hD i n th ree and a ha lf y ears . I only was ted two year s w hich is equi vale nt t o fi ve a nd a hal f ye ars of g radu atio

ears . I only was ted two year s w hich is equi vale nt t o fi ve a nd a hal f ye ars of g radu atio n. W hat can I do ? Wh at c an I do? Tha t's what I w as l ike. And my pare nts, I w as i n go od h

n. W hat can I do ? Wh at c an I do? Tha t's what I w as l ike. And my pare nts, I w as i n go od h ealt h at tha t ti me. And I di dn't hav e an ythi ng t o wo rry abou t at hom e. S o... You did wha

t yo u di d. I did wha t I did. Wha t do you hop e to be Hu Y uanm ing in t en y ears ? I stil l ho pe t o ma inta in a ll t he g ood qual itie s to day. And the n le ave all the bad qual itie s be

hind. Do y ou k now what tho se b ad q uali ties are ? I know . Wh at a re t hey? For exa mple I

hind. Do y ou k now what tho se b ad q uali ties are ? I know . Wh at a re t hey? For exa mple I can' t ge t bo red, I'm too bus y. F or e xamp le, I ca n't mana ge m y ti me w ell, I'm ver y ti

red. For exa mple th e ti me I spe nd o n my wor k is mor e th an t he t ime I th ink. I t hink the

red. For exa mple th e ti me I spe nd o n my wor k is mor e th an t he t ime I th ink. I t hink the

se a re a ll g oing to be c hang ed. Than k you.