Best of the Pod: Reid Hoffman on How AI Is Answering Our Biggest Questions

Why do you care about philosophy? Why

are answering these big questions important?

>> You know, one of the things I sometimes will tell MBA schools background in philosophy is more important for entrepreneurship than an MBA. Philosophy

is very important to this stuff because it's understanding how to think about very crisply what are possibilities, what are theories of human nature as they are manifest today and as they may

be modified by new products and services, new technology, etc. Usually in this show, we talk about like actionable ways that people use LGBT, but a more interesting question is how

does AI in general and how might it change what it means to be human? These

are really deep, big philosophical questions. I thought you might have a

questions. I thought you might have a unique perspective on this intersection.

Reed, welcome to the show.

>> It's great to be here.

>> Uh, great to have you. So, I'm sure that uh, everyone uh, listening or watching knows this, but you are a renowned entrepreneur. You're a venture

entrepreneur. You're a venture capitalist. You are an author. You're

capitalist. You are an author. You're

best known as the co-founder of LinkedIn. You're a partner at Greylock.

LinkedIn. You're a partner at Greylock.

Um, you were a board member, were a board member, um, and an early backer at OpenAI. Um, and you also have an

OpenAI. Um, and you also have an incredible podcast, Masters of Scale.

But perhaps most relevant to this conversation, uh, you also studied philosophy at Stanford and Oxford, and you almost became a philosophy professor, uh, which I didn't know before researching this interview. It's

really cool.

>> Yeah. No, it was definitely, um, part of it was I've always been interested in human thought and language. um started

with at Stanford with a a major called symbolic systems. I was the eighth person uh to declare that. Um and as a major at at Stanford and then kind of

thought, hm, we don't really know what thought and language fully are. Maybe

philosophers do. And so trundled off, you know, took some classes at at Stanford, but then also trundle off to Oxford to to see if philosophers had a better understanding of it.

>> I love it. Um it's funny. I feel like since then, symbolic systems has become the go-to like Stanford major for like curious analytical people who end up doing startups. Um, so that's that's

doing startups. Um, so that's that's pretty funny to know that you're one of the first. Um, so usually in this show,

the first. Um, so usually in this show, uh, we talk about like actionable ways that people use JGBT. Um, and and that's that's the big question. That's I think what people come here for. But

underneath that um I think what what a more interesting question is is like how does AI in general and tragic in particular how might it change what it means to be human? How might it change how we see ourselves and how we see the

world? How might it enhance our

world? How might it enhance our creativity, our intelligence, all that kind of stuff. And these are really deep philosophical questions. Um, and as

philosophical questions. Um, and as someone who uh rigorously studied philosophy and probably still thinks about those questions, I thought you might have a a unique perspective on uh

on this intersection because I think people tend to be like they're either in the philosophy camp or they're in the like language models camp and like people who are sort of in the middle is kind of kind of an interesting one. Um,

and what I wanted to start with because I think there are probably people who are listening or watching who are like why I just want Reed's actionable tips.

um uh is is to is to ask like why like tell me more about why you care about philosophy and I think you got into that a little bit in in talking about how how you got into it but like yeah tell us why is why do you care about philosophy

why are answering these big questions important. So, you know, one of the

important. So, you know, one of the things that I sometimes will will tell like um MBA schools when I give talks there is a background in philosophy is more important for entrepreneurship than

an MBA, which of course is is is startling and and contrarian. And part

of that is to get people to think crisply about this stuff because part of what you're doing as [clears throat] an entrepreneur is you're thinking about what is the way the world could be. What

could it possibly be? what is you know you know if you wanted to use you know analytic philosophy language logical possibility or something like that but it's it's it's you know kind of what is possible and and then um partially

because you know these are human activities what's your underlying theories of human nature about how human beings are now how they are kind of

quasi eternally and how they are as they as as circumstances change as they as the environment in which you know we we the ecosystems we live in change, which

is technology and in in political power and institutions and and a bunch of other things as as ways of doing that.

And philosophy is very important to this stuff because it's understanding how to think about very crisply what are possibilities, what are theories of

human nature, what are theories of human nature as they are manifest today and as they may be modified by new products and services, new technologies, you know,

etc. And and so you know obviously people tend to say oh that's a philosophical question because it's an unanswerable question you know nature of truth or you know while we all speak and

understand languages we don't really know how that works. Um and it's part of the reason why you know there was the linguistic turn in philosophy that you know Vickingstein and others were were so known for which is well maybe these

problems in philosophy are problems in language and if we understand language we'll understand philosophy. Um and you know this question around you know these these unanswerable questions but

actually in fact like science itself is full of a lot of unanswerable questions and um and it's the working theory as we dynamically improve and that's part of what the human condition is and that's

part of what actually the in-depth philosophy is. That isn't to say that

philosophy is. That isn't to say that you know the same questions today some of the same questions today in philosophy the same questions that Plato and Aristotle and even the prescratics

and other folks are grappling with truth knowledge etc. But some of the questions are also new questions and the questions evolve and part of how sciences evolve

from philan from philosophy was this question as as we res as we get to our more specific theories of and kind of uh developing the new questions that we get

to those are outgrowths and the same thing is true in building technology uh in building product and services in entrepreneurship and that's why uh philosophy is actually in fact uh robust

and important um as applied to serious questions, you know, versus the, you know, uh one of the things I wrote my thesis on uh in

Oxford was the uses and abuses of thought experiments. And you know, the

thought experiments. And you know, the most classic one is trolley problems. Um and you know, there are both uses and abuses within the methodology of trolley problems. The most entertaining of

which, if people haven't watched it, is um uh there's a a TV series called The Good Place, which embodied the trolley problem on [snorts]

a TV episode in a absolutely hilarious way.

>> That's really interesting. What Yeah.

Like what is what is the way that people tend to misuse that because I feel like trolley problems are so common in like EA discourse and people run into that a lot online. the the fundamental problem

lot online. the the fundamental problem is is they try to frame it to get to get an intuition to derive an intuition, a principle, etc. They try to frame an

artificially different environment. So, it's like,

different environment. So, it's like, no, no, it's a trolley and the trolley will either hit the the the five criminals or the one human baby and it's

default set to hit the human baby and do you throw the switch or not? And then

when you start attacking the problem, you say, "Well, how do I know that I can't break the trolley? I could just not make it continue to run." He's like, "Well, but you know that." You're like,

"Oh, so you're positing in your thought experiment that I have perfect knowledge that breaking the trolley is impossible." So in your posit experiment

impossible." So in your posit experiment work, you're positing something we never or or when we encounter, we generally think people are crazy, right? Like you

have perfect knowledge. Like why the fact do I know that I have perfect knowledge that I can't break the trolley? Um and because you know say

trolley? Um and because you know say what what is the right human response to this trolley problem is I'm gonna try to break the trolley so it doesn't hit either of them.

>> Right. [laughter]

>> That's really interesting.

>> Right. And you might even say that the that the problem is is that to say you say even you say well you per have perfect knowledge that you can't break it. You're like, "Well, okay."

it. You're like, "Well, okay."

You know, a, you don't have perfect knowledge, and b, even if you did, maybe it's still the right response. You're

trying to get me to say, do I do nothing and run over the the baby or do I do something and run over the five criminals? Like, those are my only two

criminals? Like, those are my only two options. And you're like, well, no, I

options. And you're like, well, no, I could say even if I think I can't break the trolley, that's what I'm going to try to do because that's the moral thing to do.

>> I've actually ne I've heard a lot of trolley problems and I've never heard anyone pause at the third option. I love

that. Um, [laughter]

that's great. Uh, and I also like there's something about that where it's like, yeah, certain thought experiments sort of like hijack your instincts and and you don't quite um reason through

all these all these hidden assumptions that I think honestly reminds me of like certain doomer arguments. And I don't I don't want to like go into go into the full thing, but I think it's a it's a really interesting uh way to think about

it. If I had to like summarize what what

it. If I had to like summarize what what you just said, like the value to you of philosophy is like um thinking crisply, thinking crisply about possibilities, thinking about um human nature and

reality, all of those things are like really really really important for business people. Um I I want to kind of

business people. Um I I want to kind of like take it take another step which is like some of those some of those questions that philosophers like uh or philosophy students or philosophy nerds

just like sharpen our skills on. there

are so many of these some of these big questions um some of the big perennial questions um like what is truth what is reality what can we know all that kind of stuff I'm kind of curious if you have

a sense um as we start to get into talking about AI stuff um what are those questions where um AI large language models are are going to give us a little

bit of a new lens on on on some of those questions or what are what are questions where we'll we'll find new ones to ask that are better than previous ones even if they maybe don't answer them. Do you

have a sense for that?

>> Well, I mean historically it's like for example questions that have led to, you know, a bunch of the sci scient various science disciplines, right? It's, you

know, everything from things in the physical world to things in the biological world like germ theory and all the rest. Um I think it's actually even true. It's one of the reasons why

even true. It's one of the reasons why kind of philosophy is the root discipline for many other disciplines.

when you get to questions around like okay you know how do you think about economics and game theory or how do you think about um you know kind of um uh you know kind of political science and

real politique and and and kind of the conflict of nations and interests and it's also one of the reasons why you know as a you know probably one of my

deepest critiques of the non-reinvention of the university is the intensity of disciplinary arianism. Um so you know

disciplinary arianism. Um so you know it's just the discipline of just you know political science or just the discipline of even philosophy as opposed

to multiddisciplinary um you know and if I you know part of the thing that I tend to think is kind of an interesting thing is how much the academic disciplines tend to be more and

more disciplinary versus the hey you know maybe every 25 years we should think about blowing them all up and reconstituting them in various ways. Um

and that would be actually a better way of thinking and why some of the most interesting people are the people who are actually blending across disciplines

um within academia. And I think that that part of it is I think uh extremely important and part of the question in philosophy is the kind of the question of like well how do we evolve the question of what do we know and

obviously you evolve the question what you know through like for example a lot of the history of science is instrumentation you know new new measurement devices um that that help

with kind of you know kind of provisioning of theories um but also and that's one of the reasons why like people frequently don't think enough about how technology you

uh helps us change what is the definition to be human. Um because we have this kind of imagination, you know, like the decartian imagination that we

are this kind of this pure thinking creature and you're like, well, if we've learned anything, that's not really the way it works, right? Um that doesn't

mean that we don't think that way to have abstractions to generate logic and theories of the world and all the rest but you know um you know u you know put

your philosopher on some LSD and you'll get some different outputs. [laughter]

>> That may that makes sense. Um, so I I guess like like along those lines, if you if I step step back and squint, I can kind of like uh you can kind of

divide the history of philosophy into um essentialism and nominalism for for a certain part of philosophy, right? like

um and essentialists are like um do you believe that there are like fundamental there's a fundamental objective reality out there that's knowable and that there's a way to kind of like carve nature at its joints and nominalists um

which where we would include Vickenstein which I know I know you you studied pretty deeply um and pragmatists um uh think that more or less truth is is more or less relative or it's about social

convention or it's about what works or there's a lot of different formulations of it and there's this sort of like ongoing debate between people who think one thing one thing or the other. Do you

think language models like change um or add any weight to either side of that debate?

>> I think they add perspective and color.

I don't think they resolve the debate.

Um the and there's certainly some question about since they function more like later Vickingstein or more you know kind

of nominalist um you know you say well does that does that weigh in on the side of nomalists because of actually in fact the way they function and actually in fact you say

well if you look at how we're trying to develop um the large language models we're actually trying to get them to embody body more essentialist characteristics as they do it like how

do you how do you how do you ground in truth have less hallucination you know etc and you know to to to gesture at a different uh earlier German philosopher

you know Hegel one of the things I'm uh I think is kind of part of a I think as kind of the human condition is that thesis antithesis synthesis like

you could say hey we have an essentialist thesis we have a nominalist antithesis And the synthesis is how we're putting them together in various ways because

you say look we uh and I don't even think later Vickenstein would have said that the the world is only language you know kind of what you know the deconstructionist and Terry Dah went to

was like you know it it it is only the veil of language and you have no no contact the world and you so you're not grounded in the world at all. I think he would he would he would think that's

kind of absurd, [laughter] right? But

his point was is to say that there is also in how we live as forms of life the way that it operates is not a simple you know kind of denote of and he understood

it wasn't just denoting the cat on the mat or the possibilities the cat is on the mat and the possibility the cat is on the mat but actually possible configurations of the universe and there was this kind of notion of logical

possibility that was described as one one language of possibility was to say that kind of essentialist about a language possibility is actually incorrect to actually how we discover

truth and how we operationalize truth.

And you still have a robust theory of truth, which is not essentially what the deconstructionists do, but the robust theory of truth is partially grounded in

this notion of language games and a and a biological form of life um of how you do that. And then obviously you go into

do that. And then obviously you go into this deeply with saying well okay how is mathematics a language game as a classic language of truth is a way of trying to understand that and that's part of where

you get what philosophers refer to as krypkinstein um you know the Saul krypkkey excellent you know uh lens on reading of part of what vickingstein was

about and you kind of then apply all that you know everyone's going where is this going to large language models uh and you say well Actually in fact uh you know language is this play out of this

language game. Large language models are

language game. Large language models are playing out this language game in various ways. But part of what is

various ways. But part of what is revealed is we don't just go truth is what is expressed in language. Truth is

a dynamic process and and kind of human discourse could be synthesis antithesis you know you know thesis antithesis synthesis or other things is this human

discourse that's coming out of out of um you know this dialogic period this truth discovery this logical this this this reasoning whether it's induction is

reasoning whether it's you know um uh you know abduction whether it's you know deduction and you know these reasoning processes that get us to what we think are these kind of theories of truth that

are always to some degree works in progress.

>> That's that's really fascinating. I I

want to try to summarize that in case um in case it was a little bit difficult to follow to be honest. Like there's a there's a point in there that I think I missed something. So you tell me what I

missed something. So you tell me what I what I missed. But I think one of the like some of the things that I heard in there that that I I thought was really interesting is um uh when you think

about how we built AI which is predicting the next token that's a very um sort of late Vickenstein compatible um idea or prag pragmatic like

compatible idea where it's really about the relationship between different words in a sentence and it's not we're not finding anything out about the world like there were other AI approaches I don't know in the 80s or 70s where it

was like literally like let's list out every single object in the world and those didn't really work. Um and that would be like something along the lines of a more essential approach to AI. Um

and um uh the one that works is is is a more pragmatic a more late vicinian one.

Um but um what's what's what's quite interesting is now that we're we've we have that pragmatic base that we've bootstrapped, we're in this process of

um trying to make it more grounded more grounded in reality or more um uh more uh reduced down to like being able to talk about the essential ground

truth. Um, and I think what's really

truth. Um, and I think what's really interesting about Vickenstein is he's sort of famous for saying like the the limits of my language are the limits of my world. I don't know I don't remember

my world. I don't know I don't remember if that's um late or early. Um

but but more or less like I think what you're saying is that Vickenstein doesn't think that like we there's nothing outside of language, but he does think that the way we talk about the world um or the way that we use language

is part of this sort of like social discourse where we're all kind of like um going back and forth to like co-invent um language and structures and language games together. Um and and you

kind of see that happening with language models where like when you do something like RL RHF um like that's sort of us playing with a a language model like a playing a language game to be like no no you don't talk up

you don't talk like that is is that like generally what you're what you're getting at.

>> Yes. Um so everything you said but then the additional thing which you know uh later Vickenstein was really trying to explore in various ways because he

wasn't trying to do a kind of a completely just social construction of truth. Um, you know, I'm I'm a I'm

truth. Um, you know, I'm I'm a I'm actually a fan of you have to be a Vickingstein scholar to actually understand how both early and late Vickingstein are actually part of the same project. And late Vickingstein

same project. And late Vickingstein wasn't early Vickingstein was an idiot and now let me like I've religiously converted to this different point of view. But there is a a particular thing

view. But there is a a particular thing which is how do you get to the notion of understanding truth? And truth is is

understanding truth? And truth is is that dynamic of of discovery through language and through kind of it it has

to have some explicit external conditions that it isn't my truth your truth there is only to some degree our truth or the truth in various ways and

how do you get to that as what you're doing and having you know truth conditions and in kind of early Vickenstein the truth condition was it cashes out into to a state of

possibilities and actualities in this logical space of possibilities which include physical space as part of but broader than that. And then um later

Vickenstein said well actually in fact this modeling of logical possibility is actually not the fact the way this works right and we're not actually in fact

grounding it that way. The way that we're grounding it is in the notion of of how we play language games, make moves in lang language. And the way

that's grounded is to some degree sharing a certain biological, you know, kind of um form of life by which we recognize that's a valid move

in the language game. This is not a valid move in the language game. Now,

this is what's interesting when it gets to large language models because you go, "Well, large language models, are they the same biological form of life as us or are they different? And how does that play out?" And I think Vickenstein would

play out?" And I think Vickenstein would have found that question utterly fascinating. Um, and really would have

fascinating. Um, and really would have gone like very deep on it trying to figure that out because um, and by the way, the answer might be sum and sum,

not 100% or 100% no. 100% yes, 100% no.

Because you know the argument in favor is the large language models are trained on the corpus of human knowledge and language and everything else and they're doing language patterns on that. Some

might even argue that some of their patterns are very similar to kind of the patterns of human learning and brains.

Others would argue that it's not. Um but

then you'd say well but it's also not a biological entity and it's it it learns actually very differently than human beings learn. And so maybe its language

beings learn. And so maybe its language game which looks like it's the human language game is actually different in significant ways. And so therefore the

significant ways. And so therefore the truth functions are actually very different. And in a sense what we're

different. And in a sense what we're trying to do when we are modifying and making progress with how we build these LLMs is to make them much more

reliable on a truth based. Like we want we love the creativity and the generativity but we want it to uh to almost for for a huge amount of

the really useful cases in terms of amplifying humanity. We wanted to have a

amplifying humanity. We wanted to have a better truth sense, right? I mean like the the the the paradoxes in current GBT

um are when you know you can kind of tease it out with like very simple questions around prime numbers and you go well you you know you got that answer wrong is oh yeah I got it wrong here's the answer like well that answer is

wrong too oh I got that one wrong too here's the answer and you know a human being understand these things I'm just getting these things wrong but like [laughter] I got it like I get I'm I'm

wrong as opposed to, oh, I'm sorry, you're right. I got it wrong and here's the

right. I got it wrong and here's the other here's another wrong answer.

Right.

>> Um, and and we're trying to get that truth sense into it as a way of doing because we do have some notion of of, oh, right, this is what's characteristic

like like mathematics get gets us in a very pure definitions of of certain kinds of language games. It's one of the reasons why you know you know centuries

ago people thought math was maybe the language of the universe or language of God or language of etc because you're like okay there is the one where the purest truths some of the purest truths

that we know 2 plus 2 equals 4 is kind of embedded in and we're still working that out as we play with how we create these language tools these language

devices and it's part of the reason why I think this question is really interesting because you can actually model it to some of the the actual as it were the technological physics that

we're trying to create when we're doing the next version like like like how do we how do we get these things into good reasoning machines not just good

generativity machines and they have some reasoning from their generativity but like part of the the classic showing where they break is showing where their

reasoning stops working in ways that we value and aspire to in terms of what we try to do at at as human beings as uh in

our best selves. Here's something

exhausting, the export import dance. Let

me know if this sounds familiar. You

design something in Figma, you export it, you paste it somewhere else, and you pray that nothing breaks and usually something does. It's almost 2026, which

something does. It's almost 2026, which means that it's definitely time to stop doing this. Framer already built the

doing this. Framer already built the fastest way to publish beautiful productionready websites, and now it's redefining how we design for the web.

With the recent launch of design pages, a free canvas-based tool, Framer is more than a site builder. It's a true all-in-one design platform. From social

assets to campaign visuals to vectors and icons, all the way to a live site, Framer is where ideas go live from start to finish. Framer's design tool is

to finish. Framer's design tool is different from old school website builders you might see advertising on other podcasts. It offers vector

other podcasts. It offers vector editing, 3D transforms, gradients, animations, all for free. It has

unlimited projects, unlimited pages, and unlimited collaborators. But what really

unlimited collaborators. But what really changed how I think about Framer is that there's no handoff. What you design is the website. No developer

the website. No developer interpretation, no can you make it match the mockup conversation. You design it, you publish it, and it's live. Are you

ready to design, iterate, and publish allinone tool? Start creating for free

allinone tool? Start creating for free at framer.com/design and use the code

at framer.com/design and use the code dan for a month of Framer Pro. That's

framer.com/design

promo code Dan. Rules and restrictions may apply. And now back to the episode.

may apply. And now back to the episode.

That's really fascinating. You said a lot there. Um I really want to get into

lot there. Um I really want to get into the reasoning thing in a second, but I want to go back to the um the sort of the way that you talked about late Vickenstein versus early Vickingstein because I haven't really heard it said

that way. And the usual like thing

that way. And the usual like thing people say is like he just disagreed with everything when he was older or whatever. And what I hear you saying now

whatever. And what I hear you saying now um is more or less uh in both like in both cases he he's

saying some of the some of the same things or he has some of the same views but like the real difference is how he cashes out what it what it means to be true something is whether something is

true. And in the first um in his like

true. And in the first um in his like sort of first period he's uh talking about truth in terms of um a logical space of possibilities um that can be

broken down into these like little what he calls atomic facts and those are never really defined but like um you can kind of build up truth from there um uh mapping those those possibilities into

actualities like what's actually in the world and in later Vickenstein it's all about um these sort of like language games, the social relationships, like the the use of that word or that phrase

um in the context of people. [snorts]

And one of the things that I I really wanted to ask you about is like that first that first version of Vickingstein [snorts] is uh where it's sort of that logical space of possibilities.

Like what that reminds me of is embeddings.

um where you know embeddings are they're one of like the key underlying um uh technologies that that gave rise to AI

right in in like traditional NLP they're like allowing you to represent um words uh or tokens in a high dimensional space um and then the language model like innovation is kind of like it's not just

words it's words in their particular context each each word in a particular context has its own part of the space so um in a in a language model the word

king if if it's if it's tokenized that way um you know there's a king in chess there's a king there's an actual king there's like a king of England there's a king leer and they're all kind of like kings but they're like different spaces

and language models are able to represent um all of those different like when when we say king we mean many different things they're able to represent all that and that just

actually reminds me a lot of of like atomic facts or or or the the first like Vic Vickenstein's uh early early work and I'm just kind of curious like because I think you said that language

models sort of because of the next token prediction they they're they're sort of late Vickensteinian but I wonder how you like factor in the fact that embeddings work and they're sort of a core part of this.

>> Well and actually this is part of the fact that late Vickingstein is not early Vickenstein was an idiot. Um because

yes, I do think that the kind of the notion of call it as it were a probabilistic bet for what are the set of different

tokens that apply are are kind of there. Now the the reason why I would kind of slant more as current practice late Vickenstein than

early Vikingstein is because early Vickingstein thought that once you had the the grasp on the logic of it, you

then almost by speaking correctly couldn't make truth mistakes because the logic was embedded in it. And um and

even though the token embeddings are kind of you know part of a very broad symbolic you know quasi symbolic I would

say you know kind of network. Um and the reason it's quasi symbolic is because it's still kind of activations and so forth and isn't you know p purely the

reasoning around a token of king or you know 15 different tokens of king or 23 different partial v tokens of king. As

much as there's kind of conceptual spaces in that tokenization as mapped from a very large large use of language, but part of language isn't isn't just

the historical language, but is the the relication of it. Like if you say this is the king of podcasts, right? Or this

is the king of microphones.

>> Not yet. Maybe. [laughter]

>> Yes. Yes. But just, you know, kind of as as instances. That's part of why, you

as instances. That's part of why, you know, kind of later Vickinstein went to well, it's how we're playing these language games and how we're reapplying them. And when we say, like, for

them. And when we say, like, for example, we say on this podcast, this could become the king of podcasts. We

all have a sense of what we're doing.

It's like, well, what what would be the cases where that would be true and what would be the cases that be false? And

what prediction is that making? And and

how is it that that's a you useful thing? I'm sure someone said king of

thing? I'm sure someone said king of podcast before, but I've never heard it before, right? And it's a different

before, right? And it's a different tokenization especially as it gets developed and elaborated a lot in in discussion. And then actually if you

discussion. And then actually if you suddenly had another you know terabyte of information about discussions of kings and kingdoms and you know all the

rest and all of a sudden that token space that it's learning from would change right u and then the generalizations off it would change um and that's part of the reason I would

say it's kind of more later vickingstein even though not completely not completely disconnected from those embeddings early And it's one of the

reasons why like actually in fact later Vickinstein is not truth is just what language says. It's no there's there's

language says. It's no there's there's ways in which it's embedded in the world by how we navigate as biological beings and that's part of how the world kind of

comes and impacts it and therefore it's not just language by itself free floating like the cartisian consciousness but it's embedded in some ways and part of what he was trying to do is figure

out well from a philosophy standpoint how do we understand those embeddings and how do we drive our truth discourse in language

based upon that biological embedding.

>> That makes sense. So I think what I what I hear you saying is um uh despite the fact that embeddings are in this sort of uh they're they're

mapping words into this high dimensional space which sort of seems like um kind of mapping words into this like sort of atomic facts or like logical possibility

space. um the way that that space is

space. um the way that that space is constructed and and what makes something go into one part of the space or another is more late Vick and Shinian because

it's very much about how it's used in practice um and whether it's useful for humans in the world rather than like it's about some deep underlying um

logical ordering where uh if you've created that ordering like you can't say anything wrong because uh because you're use you're only using words in that in in from that space. Does that is that is that kind of on target?

>> Yes, exactly. And and and part of it is we know that there is that there's truths that where the coherent use of language still is a falsity. And so how

do like part of what you we're trying to figure out is how do we get more of those truths and truthtelling and reasoning because reasoning is about

finding truth. um you know into how do

finding truth. um you know into how do these you know LLM work >> and what do you and and just just to move into that point a little bit like

do you think that um like what is most promising to you in terms of like ways that we we are getting reasoning into these language models and do you think that there are any like philosoph like ideas from philosophy whether

Vickenstein or otherwise that are relevant to that to that uh project?

Well, the answer is certainly yes on the relevant ideas. Currently, I think we're

relevant ideas. Currently, I think we're doing a couple things. So, I think we're we're taking kind of call it, you know, human knowledge and figuring out how to

get that as part of what's trained.

So the earliest discoveries um were actually in fact if you trained on code computer code then these models learn patterns of

reasoning much broader than just computer code and so all of the models that are doing this are now also training on computer code

even if they don't have a target of being a you know Microsoft co-pilot you know code generation you know etc like even if they're not doing that because there's a re there's there's a pattern

just like math of of crisp you know kind of um you know modeling of of reasoning.

Another one is that's currently happening is well what are you doing with textbooks and the notion is if you take the same kind of training

discipline that we use for human beings encapsulated in textbooks you can for example build much smaller but still very effective models based on textbooks

as ways of doing it. And so textbooks is another one. Now as you begin to like

another one. Now as you begin to like there's probably like some interesting as it were computational philosophy if you begin to say well how do we cash out

kind of theories of um you know whether it's you know kind of uh you know call it theories of science in the kind of different

theories of science and you're kind of building those models into you know how do you get you know it's kind of like lacatosh as a development on popper

given thinking about coonian you know kind of models of scientific paradigm um how do you um you know kind of make

you know kind of predictions on those kinds of bases and you know some of the indepth work in logic maybe basian logic

um as ways of possible possibly looking at this I'm quite certain that there probably are some some very useful things to elaborate beyond it. Now,

currently, of course, part of the the notion of these things are their learning machines. So, you have to have

learning machines. So, you have to have give a fairly substantive corpus of data from them to learn from. Um, now of course there's synthetic data and the

like there may be like philosophy is in what patterns do we create synthetic data that is still useful to learn from off the act the current data, you know, might be anyway. So there's there's a

bunch of different kind of gestural areas, but I'm certain those are there there even I don't even though I'm not bringing up I'm I'm I'm making gestures

rather than, you know, uh specific uh theories um as to how that they're there cashes out.

>> That's really interesting. So it seems like um basically the way that we're trying to get reasoning into models is to find sources of data that just has really crisp reasoning and so they'll

like learn the reasoning from that. Yep.

>> I'm sort of curious like uh if if that's the case like aren't there are only a certain number of like moves you can make in logic. Um you know like you can do induction, you can do deduction, you

can do there's there's like not there's not like infinitely many moves. like why

if if we have a a really crisp set of of data on that's sort of teaching them these moves, what's the like thing that's sort of stopping them from being

able to apply them more more broadly. Um

and maybe that question is not wellformed.

>> Well, first yeah correction of the question because actually in fact in logic there are infinite moves. One of

the things that's interesting in in various logics is different orders of infinity. um as people kind of think

infinity. um as people kind of think through it. So there is various things.

through it. So there is various things.

Now what you did actually remind me of is one of the things that I'm I've been recently rereading um because of thinking of Girdle's theorem as kind of a classic instance of

of human meta-inking. And so Girdle Asherbach um which I read as a high school student. I've been rereading

school student. I've been rereading recently u because I'm >> that's great. What do you think? Uh

well, it's it's it's this it's this this tangle of amazing observations that you're trying to kind of like I'm trying to think about it from a viewpoint of modern LLM. So, it's kind of like this

modern LLM. So, it's kind of like this question of you got the girdle self-reflection which is roughly speaking in any sufficiently robust language

system there are truths that cannot be expressed within the language system, right? Um, and like that's

right? Um, and like that's mind-boggling, [laughter] right? And

what exactly it means and so forth. And

it's because of this classic kind of diagonalization proof to say if you're enumerating out all the all the truths, there's at least one of them that's not

captured in your your in your in your numbering out of all truths. Hence, one

version of kind of infinity. um you get that in the recursion patterns um that you see within Echer and within within Bach that you say that's another recursion

pattern because there's a recursion pattern of getting to showing the shadow of at least one truth that's not captured within your enumeration of all the truths. um you go okay well what

the truths. um you go okay well what does this mean for thinking about truth discovery whether it's human truth discovery LLM truth discovery and that

kind of the the what are the things that are outside the boundaries of logic like it would have been um like I would have

been very curious to have Girdle and Vickenstein two folks very focused on logic to talk about Girdle's theorem like like I would have like you know I

was asked recently You know, if I had a time machine, would I want to go forward or back? Me, I'd

rather go forward. I'm very I'm just curious about how do you shape to the future? But like one of the the

future? But like one of the the historical back ones that I would love to do is put Girdle and Vickenstein in a room and say, you know, Girdle's theorem discuss [laughter]

you know, and like like like you know, I would I would do a lot to try to be able to hear that conversation. We needed we need some GBTs in here with with Girdle uh with Girdle and Vickingstein. Maybe

Girdle doesn't have enough writing to make that happen. But uh maybe maybe eventually and the twistiness of the thinking is one of the things that is that is you know is one of the things

that made Girdle so so spectacular in this you know. Uh another one by the way that were historical walks is Einstein and and Girdle used to take walks. you

know, you wish that you had digital recorders, like please record the conversation [laughter] right? We we would really like to listen

right? We we would really like to listen to that.

>> Um, no, I love that. Um, that's really interesting because I feel like like I read God in college. I loved it. The

thing that's so good about it is it's like it's such an interdisciplinary book, you know? It's got math and music and art and like all like all this stuff and you're like, "Wow, like that's the

kind of mind that's going to invent new minds." And then you you see Hoffadder

minds." And then you you see Hoffadder today and he's like sort of not like he's not definitely not in the LLM conversation. He's a little bit freaked

conversation. He's a little bit freaked out by them. Um and like I'm kind of curious like what do you what do you make of that? Like what did he get right and what do you think he got wrong?

Well, I think a central thing that he got right, at least to how I operationalize is, and that was the reason I was gesturing at Hegel with

thesis, antithesis, synthesis, which is it's a dynamic process that's ongoing and you can't necessarily predict the future synthesize and that's part of even though obviously in philosophy you

try to articulate the the truths, you know, that the cart I think for am or um you know, Vickenstein saying, well, there actually have to be a world in a certain way to to actually there to be

truth statements in the language statement of I think therefore I am and so therefore you can be you know kind of broader than just the disembodied mind

um as a way of of thinking about that because you think about what the truth conditions must be in a language if you're saying if you're saying in a way that's coherent to your current self and your future self I think therefore I am

what are the truth conditions in the language as ways of doing it and So, but that's a dynamic process by which we are making new discoveries and that's kind

of the synthesis and that's the thing that I think is is um you know is part of what I take from the

kind of the girdles orbach interweaving of these different of these different dynamics and showing the kind of the patterns across it. Now frequently when you go across a lot of areas where

people say hey we have this language system and all we know is through our language and then they kind of go and the world is unknowable to us because the only thing that's knowable to us is our language. You say well that's

our language. You say well that's presuming there's no relationship between how the language engages with the world and how we engage with the world with the language. And so it's one of the reasons why you get into really

interesting you know biologists like Varela and Matana. It's the reason why, you know, you get to, you know, kind of different patterns of self-referential logic. And so you it gets very

logic. And so you it gets very interesting and so I don't I myself don't get freaked out by LLMs and part of this. I think wow new things that we

of this. I think wow new things that we can discover. Right.

can discover. Right.

>> Right. And how does that um make the discourse much richer, much more valuable, much more compelling and in some ways uh higher ontarget you know

discoveries of the truth like because I gave a speech in Bolognia last year uh where uh along with the book I published last year impromptu his last

chapter is homoche is that one of the things that we think of ourselves as human beings as static and actually we're not static as We are constituated by the con the technology that we engage

and bring into our being. So for

example, you and I are looking at each other on this podcast through glasses.

Like think about the world with glasses without glasses, right? The world is a very very different place and how you can perceive like we say most of our theories of truth are fundamentally

based on kind of perception. Like you

know seeing is believing is kind of a classic idiom. And well, if you don't

classic idiom. And well, if you don't have glasses, how you see is very different, right? And so, so like

different, right? And so, so like technology changes our landscape in the perception of truth. You know, that's why microscopes and telescopes and and all this rest these other things that

kind of get to that changing that landscape. And that's part of what we're

landscape. And that's part of what we're doing with technology and we're doing in this particularly interesting ways with these LLMs in terms of how they're operating.

>> Yeah, that makes that makes a lot of sense. And I I love that point um about

sense. And I I love that point um about sort of how technology changes us um and and really like how flexible humans are.

It reminds me a lot actually cuz I read I read your book to prepare for this and I also I read your Atlantic article and your podcast on this like um and it reminds me a lot of Have you read the book The Weirdest People in the World by

Joseph Henrich?

>> No, I probably should.

>> It's really it's really great. He's a

psychologist at Harvard. And the the point of the book is um most of what we take to be the psychology literature is wrong. And it's not wrong because of P

wrong. And it's not wrong because of P hacking and all that other stuff, but it's wrong because um the psychology literature is based on studies of Western college students. And Western

college students have a completely different psychology than like us people everywhere else in the world um now and in history. And one of the key diffic

in history. And one of the key diffic uh in western college students is that they can read. And reading changes your brain in all of these different ways. It

enlarges parts of your brain and shrinks other parts where um for example, if you're if you can read uh you're more likely to pick out like objects in a landscape rather than see like the

holistic uh the holistic scene. And

there's a bunch of these other like significant differences that you find in humans who can read versus humans who can't. And so like reading as this

can't. And so like reading as this technology uh created all this stuff like it um you know one of one of the things that he he argues is that

uh it allowed us to create uh like a society where we had um uh where we had churches that that created like rules and principles that like people would follow even though they weren't being

watched. So, like you know, you know,

watched. So, like you know, you know, I'm not supposed to like steal or whatever. And you you can't it's like

whatever. And you you can't it's like really hard to get like uh big organized society without without reading basically is is like one one big point

of of the book. And that it's because it changes our our actual biology. And I

think that's the thing that um that people sort of miss about language models. Like not to say that like we

models. Like not to say that like we should ignore that there are like any any language models dangers or anything like that. Like there's a lot of I think

like that. Like there's a lot of I think really interesting and really important problems to solve, but like um when you think about what language models might replace versus augment,

I think it's also really important to like know that like we've been replacing or augmenting ourselves for many many many many generations. And um if you took a human from like you know five

generations ago or 10 generations ago and put them put them now like it would be like really hard for them to like interact in our society now. Same thing

if if you took one of us and pushed us back in time. Um, and and that's because like we we sort of like uh we grow and change in response to our environment and our culture, which is like this

collective memory that like that gets loaded up so that we're a modern human instead of like a pre-evolutionary human or whatever. And the same thing is going

or whatever. And the same thing is going to happen with with language models, like you can kind of like put it on this on this uh timeline from the invention of language to like reading to the printing press. Like it's all the same

printing press. Like it's all the same kind of cultural transmission technology. I've I've heard some

technology. I've I've heard some researchers call it and I think that that that's exactly kind of like what it is to me. Um curious what you think about that.

>> Well um you know I I definitely think that the progress of cultural knowledge um and I don't know if it's the same author but the secrets of the secret of

our success um is is I think a very good book. Um and you know it's partially

book. Um and you know it's partially because how we make progress is updating our cultural knowledge. It's part of the reason why it's not surprising that then when we we generate uh interesting

learning algorithms that we can apply to the human corpus of knowledge that we then generate interesting things that come out of that because that's

essentially a a a partial index of cultural knowledge. It's not the

cultural knowledge. It's not the complete index because as you know like for example the secrets of success go through it's like well you know how do you identify which things to eat or which things not to eat or when to do

that and all the rest of that and that's part of how you make progress and I think that's essential part of how we're um how we actually evolve like everyone

tends to think evolve in human beings is you know do we evolve to be faster longer stronger genetics and actually in fact a major clock of our evolution is

we shifted like you could say there's geological evolution which is super slow then there's biological evolution which is slow and then there is cultural

evolution or knowledge digital etc which is much much faster and part of how the kind of the secrets of our success is we we is we got into kind of cultural

evolution and um you know and kind of that progress of digital and that part of what we're doing with AI I and LLM is

tools to help accelerate that you know cultural digital evolution which can include like why is everyone going to have a personal assistant because the personal assistant will be I read all

the texts and I can bring them to you as as as as you're talking and trying to solve problems. So like for example on the you know what are things that people

should be using chat GBT for is obviously a immediate ondemand personal research assistant that today hallucinate sometimes and you have to be aware of that and kind of understand

that but an immediate research assistant is one of the things that is obviously here already today. Um and and you know if you don't think you need a research

assistant it's because you just haven't thought about it enough. [laughter]

Yeah, I mean it's it's it's incredible.

Like it it it takes the every everything that humanity knows and gives it to you in the right context at the right time when you ask for it. And that's exactly

kind of like the bottleneck of cultural evolution is like getting the right information out to the edges of people that need it instead of like having it be locked up in a in a on the internet or like in a library or whatever where

you have to go expend resources to get it. like all those are better than

it. like all those are better than having to transmit knowledge orally for example. Um but uh but yeah, language

example. Um but uh but yeah, language models are like a a profound next step.

Um so we're we're getting close to time. I

have a couple I have a couple of uh we we had a whole final section about science, but we may not be able to get to science. We'll have to maybe do a

to science. We'll have to maybe do a part two.

>> Yep, that'd be great. I'd be up for that. I love these these topics. And um

that. I love these these topics. And um

but I want to ask you a couple a couple more things like just sort of on on the you know philosophy in AI in AI uh front. So like

front. So like why do you think philosophers didn't come up with AI like why did it why did it come out of I mean I guess it it came

out of like sort of a computer sciency uh tradition but also just like really sort of an engineery people who just were making stuff. Um uh yeah, talk to

me about like why why it didn't come from philosophers.

>> Well, I do think that this is a little bit like I was gesturing at earlier which is being disciplinarian

is I think um you know has obviously people are not idiots in doing this.

they have some strengths in it but also some weaknesses. And um you know and I

some weaknesses. And um you know and I think part of it is to think about like well how is it that technology is going to change our conceptions of how we use

language and how we discern truth and how we argue about it and all the rest of the stuff is I think um you know pretty central and you know it's kind of like you know

how is technology as ways of knowing or ways of perceiving or ways of communicating or ways of reasoning important and you know philosophers will

say you don't need any of that we just I sit down and I cogitate you know kind of [clears throat] a [laughter] you know kind of canonically you know de cart um

and look I think there's a there's a role to sitting down and cogitating but I think there's also a role to discourse and it doesn't necessarily

mean you have to be a externalist or you know a you know a kind of I don't know who the physical materialist, you know, um, you

know, advocates are, you know, the the church lens and other people, um, you know, back in the days when I was a philosophy student were those among

those who were who were, uh, very vocal on that. Um but is to say that actually

on that. Um but is to say that actually in fact this notion of how do we engage technology in our work is a very good

thing to do and if so then maybe philosophers would have come up with it more or would have been able to participate more in it versus the you know computer scientists who are like okay I'm working on the technology side

of it what can I make with this technology and obviously you know the what can I make with this technology goes Well, earlier than computer science, right? I

mean, you've got you go all the way back to Frankenstein, you know, and kind of thinking about, you know, kind of imaginations about what could be constructed here or the Golem or or

Talos in Greece. Um, and so the the notion that things could be constructed now, could they be constructed with silicon and they could be constructed with computer science? You know, that's the modern kind of artificial

intelligence. But the but the notion of

intelligence. But the but the notion of that is I think one of the reasons why I want philosophy to be broader in its instantiation.

You know, not just a question around, you know, this this is obviously a bit of a deliberate rhetorical slam, but trolley problems. [laughter]

>> Yeah, that makes sense. Um maybe maybe a way to frame that is like uh it's better to be like asking deep philosophical questions and be a philosopher out in

the world to some degree than it is to just be a philosopher. I don't know if you'd agree with that, but like something something like that.

>> I chose that with my own feet.

>> Yeah, there you go. [laughter]

Um yeah, I I I I definitely I definitely agree with that. Um so so we have a minute left. Uh the the last thing I

minute left. Uh the the last thing I want to ask you is I assume that there's there are a lot of people who are listening to this maybe are not um have not been philosophically inclined in the

past and are either like wow I could not follow any of that and I want to figure out what they said or they're like oh my god like I want to learn how to think like that. Um and I think for the first

like that. Um and I think for the first group of people I would totally recommend like just use chatbt talk to chatbt about this stuff and it will tell you for sure.

>> Yes. Uh I but I wanted to ask you like if people are thinking about like they want to get that kind of like thinking crisply about possibilities thing that you that you talked about so well at the beginning like where would they start or

what are your like what are your favorite kinds of philosophers or kinds of books like this to dive into? Well um

you know I think um the best way is to get uh interactive um it's part of the reason like like study philosophy or even you know even for the second part of the question some use of chat GBT

also very helpful there because the interactive is is is what does it and like for example one of the the things that I use chat GBD for which is part of this is I have something that I'm

arguing for thinking about arguing for and I write put in my argument And I say, "Okay, Chad JVD, give me more arguments for this. How would you, you

know, argue u for this differently or more or and then also how would you argue against it, right? What would your counterarguments be to this?" And use

that as kind of again, you know, the kind of thesis and synthesis trying to get to synthesize in this. And

uh and so I think that dynamic process is really important.

Um, and so, you know, part of the the way that people traditionally try to get to this is they they go try to go through

what are some of the real instances of great human thought and then try to understand that how to think that way.

So one of the things that was too much text prompting to go into impromptu but is I think very useful as another

utility for you know kind of use of chat GBT is um you know uh like I like I'm a non-mathematical college graduate

explain Girdle's theorem to me you know I'm a non-physicist explain Einstein's thought experiments around relativity to me, you know, etc. And that dynamic

process of getting into understanding those things is part of how you learn to think this way. And it's one of the reasons why, you know, kind of our um,

you know, one of the things that has helped us accelerate our cultural evolution, our c cultural evolution, the secret of our success is having things like books, having things like

universities because it's that dynamic process of engaging that's so important.

And so there's not necessarily one specific book, although by the way, if you really want to have your mind boggled, go re, you know, read or reread Girdle Asherbach. It's great,

Girdle Asherbach. It's great, >> right? you know, but but but like what

>> right? you know, but but but like what are the instances of these canonical amazing pieces of thinking and then you know kind of in that dynamic engagement process you're internalizing them.

>> Yeah, be curious about great ideas and and engage with them. Um this was this was a great conversation. I really

appreciate you coming on it. I feel like I learned a lot. Thank you so much.

>> My pleasure. Awesome.

>> [music] >> Oh my gosh, folks. You absolutely,

positively have to smash that like button and subscribe to AI and I. Why?

Because this show is the epitome of awesomeness. It's like finding a

awesomeness. It's like finding a treasure chest in your backyard, but instead of gold, it's filled with pure unadulterated knowledge bombs about chat GPT. Every episode is a roller coaster

GPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat, craving for more. It's not just a show, it's a journey into the future with Dan

Shipper as the captain of the spaceship.

[laughter and snorts] So, do yourself a favor, hit like, smash subscribe, and strap in for the ride of your life. And now, without any further

your life. And now, without any further ado, let me just say, Dan, I'm absolutely hopelessly in love with you.