AI News: 5 New Models Dropped This Week!

So, it's been a big week in the world of AI with a ton of new models that came out, some fresh drama, and some really interesting stuff that you're probably going to want to know about. I'm not

going to waste your time. Let's just get right into it. If you're a fan of Anthropic and Claude, we got a brand new model this week in Sonnet 4.6. Now, this

model isn't quite as smart as their state-of-the-art Opus model, but it is now the best of the Sonnet models. And

if you're not paying too close of attention to what Anthropic and Claude's models are for, well, the Sonnet model is what you will get in the free and the lower tier plans. We can see here for

those on our free and pro plans, Claudson 4.6 is now the default model in Claude.ai as well as in Claude Co-Work.

Pricing remains exactly the same, but you now have a smarter model if you're using Claude. 4.6 also has a 1 million

using Claude. 4.6 also has a 1 million token context window in beta. However, I

don't believe that huge context window is available inside of the actual Claude.ai chatbot. That's more for API

Claude.ai chatbot. That's more for API users. Basically, they're saying here

users. Basically, they're saying here what you used to have to go to Claude Opus like 4.5 for now are Claude Sonnet 4.6 is just as good but available in the cheaper plans and also available for

cheaper in the API. One of the big leaps for this new model is its ability to actually use a computer. There's a

benchmark here called OSWorld which shows how much better these models are getting at being able to actually use a computer and do tasks on your behalf.

And we can see that Sonnet 4.6 has become very very capable at that. Here's

the benchmark comparison for this model.

One thing I like that Enthropic does is that they do compare it against competitor models like Gemini 3 Pro and GPT 5.2, which by the way, spoiler, we got a new Gemini model, so this is going

to be slightly different, but we'll get to that in a minute. So if we look at the Sonnet 4.6, this is the new model which is available in all the plans now.

And if we compare it to their most state-of-the-art model, Opus 4.6 here, we can see that the numbers are actually pretty close. If we look at Agentic

pretty close. If we look at Agentic Coding from SWEBench verified, Sonnet 4.6 scores a 79.6 where Opus 4.6 and 4.5

are only 80.8 and 80.9%. Agentic

computer use 72.5 compared to Opus, which is the state-of-the-art at 72.7.

Same here with this Agentic Tool use benchmark. The state-of-the-art model is

benchmark. The state-of-the-art model is 91.9% versus 91.7%.

So, the difference between their Sonnet 4.6 less expensive, widely available model is pretty on par with their state-of-the-art model at quite a bit less expensive to actually use. Now, I'm

going to be totally honest with you.

Most people that are using Claude on a day-to-day basis are probably not going to notice a huge difference in how it was performing before and after this update. This update is actually probably

update. This update is actually probably more valuable for people that are using the API because they'll be getting almost opus level coding and agentic tool use at sonnet pricing inside the

API. And there are actually a few

API. And there are actually a few benchmarks here where sonnet is state-of-the-art and even beats out opus 4.6 agentic financial analysis and

office tasks sonnet 4.6 is now the absolute best at. So if I jump into claude.ai AI here. Pretty much everybody now has this Sonnet 4.6 option. Again, I

don't really know how to test it and demo it because it's not going to feel a whole lot different to most people.

That's also the reason I didn't make a whole dedicated video making a big deal out of Sonnet 4.6. It's really cool that we have a model that's smarter, better, faster in the free and lower tiered

plans. But again, the people who are

plans. But again, the people who are going to notice these changes the most are the ones that are using it for programming and tapping into the API because now they're getting almost Opus

4.6 level quality at sonnet level pricing as well as that giant million token context window. Enthropic also

released a bit of a quality of life update with this increased web search accuracy and efficiency with dynamic filtering. Basically, when Claude goes

filtering. Basically, when Claude goes and searches the web for you, it'll often pull in an entire website, which uses up your tokens in your context window, potentially costing you a lot more money. Well, now what it's going to

more money. Well, now what it's going to do is it's going to essentially read the website, find what's important, and only pull the important stuff into context, which will save on some of the API costs

when using web search. So, again,

another feature that developers will probably like, but if you're just a regular CLA user, you're probably not going to notice a whole lot of a difference. Now, some things that might

difference. Now, some things that might genuinely be useful to most people here are the fact that Claude is now in PowerPoint if you're on the pro plan,

which I believe for Claude is the $20 a month plan. So, we can see a screenshot

month plan. So, we can see a screenshot here of what Claude in PowerPoint looks like. I don't actually use PowerPoint

like. I don't actually use PowerPoint myself, so I don't have it installed on this computer, but we get a little Claude sidebar where you can prompt it to do things in PowerPoint as well as analyze things in the PowerPoint. I

think the two features that excite me the most about this if I did use PowerPoint more often would be the ability to generate a full deck from a description because I hate starting anything from scratch as well as

creating native charts and diagrams based on data that I feed it. That to me seems like it would be really really helpful. Now, in spite of the fact that

helpful. Now, in spite of the fact that Enthropic rolled out an entirely new model this week, I think the coolest thing that I saw from Anthropic this week was this in collaboration with Figma from Claude Code to Figma turning

production code into editable Figma designs. So, we can see in this demo

designs. So, we can see in this demo here, they say spin up a dev server and it opens it up on their local host and then they say send the my pantry page to the Figma file. It thinks for a second

and then it takes the design that's there on the site right now and tosses it into Figma. And once it's in Figma, they can then edit the design. They're

working together to actually add notes and design collaboratively inside of Figma. Here we can see they redesigned

Figma. Here we can see they redesigned it into multiple rows. And then it says here they can roundtrip it back to code.

Now, for whatever reason, they didn't show that part in their demo videos, but with the Figma MCP server, you can bring work back into the coding environment using a prompt and a link to the Figma frame. So once you have the design

frame. So once you have the design dialed in the way you like it, you grab a frame, take it back to Claude Code and then it tweaks the code for you and makes it the design that you designed it in Figma. To me, that's actually pretty

in Figma. To me, that's actually pretty cool. But Enthropic wasn't the only

cool. But Enthropic wasn't the only company that rolled out a new model this week. Speaking of AI news, I've been

week. Speaking of AI news, I've been feeling like a lot of the actually useful AI tools sometimes get buried under the flashy stuff or the big names.

That's why I want to call out Warp and their new cloud coding agent platform called Oz. It's not flashy and it won't

called Oz. It's not flashy and it won't turn you into a hyperrealistic AI avatar of a celebrity, but it'll save you time and make your life so much easier. Oz is

built for professional developers who are running multiple AI agents at once.

So your agents are fixing bugs, shipping features, and maintaining code bases, but they're starting to hit local limits. Instead of running agents on

limits. Instead of running agents on your own machine, Oz lets you spin them up in isolated cloud environments and manage them from a single panel. If

you're a developer building real products and workflows and not just flashy demos, this is the kind of thing Oz is built for. And 97% of the code diffs generated by warp agents are

accepted by users, saving developers about an hour or more every day. It was

built by developers for developers with over a million lines of Rust and optimize for massive code bases. So if

you want AI agents that work with you instead of getting in the way, Oz is worth checking out. is free to try, but for a limited time, use code wolf to try

Warp Build for only $5. Check the link in the description box. And thank you so much to Warp for sponsoring this portion of today's video. Google shipped Gemini

3.1 Pro. Now, similar to Anthropic, it

3.1 Pro. Now, similar to Anthropic, it is a fairly marginal update, but it is rolling out this week, and we can see for developers, it's going to be in Google AI Studio, Gemini CLI, Google

Anti-gravity, and Android Studio. For

enterprises, it's going to be in Vertex and Gemini Enterprise. And for

consumers, we're getting it inside of the Gemini app and Notebook LM. And

similar to Anthropic, Google actually does show it against their competitors.

So, we can see Opus 4.6 and even Sonnet 4.6 here, as well as GPT 5.3 Codeex.

This is the new state-of-the-art model for Arc AGI 2, which has a bunch of abstract reasoning puzzles. And we can see that it's actually topping this benchmark here by a pretty wide margin

compared to the rest of these here. Opus

4.6 is really the only one that comes like anywhere close. When it comes to scientific knowledge, it's beating everybody else. When it comes to

everybody else. When it comes to terminal bench, it's beating everybody else. Scientific research coding,

else. Scientific research coding, agentic tool use, we were just looking at how Sonnet was the new best. We can

see their numbers here. Well, when using tools, Gemini 3 Pro is now pretty much state-of-the-art, matching Opus 4.6. So,

it's a model that actually did get a lot better, but in specific areas like scientific knowledge, agentic terminal coding, scientific research coding, and things like that. If you're a coder,

you're probably still using Opus 4.6 or GPT 5.3 codecs because those ones are pretty much the ones that are kicking everybody's butt in the coding world.

Now, one of the examples I've been seeing a ton of for showing how good this is is the ability to create animated SVGs, which we can see on the screen here. They look like pretty solid

screen here. They look like pretty solid looking motion graphics and really really impressive that it can code this kind of thing specifically. So if you are coding up a new software or a new

website or something like that, I think most of us that are doing a lot of vibe coding kind of stuff will jump around between models because different models are good at different things. And it

seems like if you want to create any sort of SVG animation within whatever you're building, Gemini 3.1 Pro is going to be your go-to for that kind of thing.

So, as of this recording, the best way to go and test out the Gemini 3.1 Pro is to go to Google's AI Studio. AI

studio.google.com here. Go to the playground up here when you select your model. Click here and you should be able

model. Click here and you should be able to go to Gemini. Gemini 3.1 Pro Preview.

Select this. And this is how you're going to be able to test it. Since the

animated SVG thing is the thing that I keep seeing, let's test that. Create an

animated SVG of a greywolf playing basketball. make it viewable in my

basketball. make it viewable in my browser. Let's run that. See what

browser. Let's run that. See what

happens. I can click here to expand to view the model's thoughts. So, we can actually see it thinking through the problem as it's creating. And after

about 3 and 1/2 minutes or so, we've got something that we can take a peek at.

Let's go ahead and click on our preview button and see what we got. Show

preview. Oh my gosh. I mean, yeah, it's it's not bad. It's actually you can tell what it's supposed to be. Some of the stuff's in the wrong position. Like the

number should be more on the back here.

The headband's kind of covering the eyes, but honestly, like, it's not horrible. Here's some other examples

horrible. Here's some other examples shared by Google themselves of a pelican riding a bicycle, a frog on a penny farthing bicycle. I didn't know that's

farthing bicycle. I didn't know that's what that was called, a giraffe driving a tiny car, an ostrich on roller skates.

They're showing Gemini 3 Pro on the left, Gemini 3.1 on the right, and we can see that it's quite a bit of a step up in improvement. I think one of the biggest differentiators is that it tends

to do a lot more gradients than it used to, which makes it look a little more impressive, but overall pretty cool.

Now, again, both of these models that came out are impressive in certain areas. Sonnet is a really good model at

areas. Sonnet is a really good model at coding for less expensive cost of Opus 4.6, but it's not going to be quite as good at coding as Opus 4.6. Gemini 3.1

is an improvement in a lot of science and research kind of tasks. But if I'm being totally honest, both of these models, I would say the general everyday user of these tools is probably not going to notice like a huge huge leap

from them. But that's not all Google

from them. But that's not all Google cooked up this week either. They also

rolled out their LRA 3 model, which is their music generation model. This is

like Google's answer to Suno. Now, it is actually pretty good, but you can only generate 30 seconds right now. And you

just use it in the Gemini app. If I jump over here to gemini.google.com,

you can see I now have a button that says create music. And also, if I click on my tools dropdown, the create music button is here as well. You can pick a track to remix here if you want. Or you

can simply describe the track you want.

For example, I'll do an upbeat dubstep song about the San Diego Padres's. And

here's what it made FOR US.

>> PETCO PARK IS GOING WILD TONIGHT. SEEING

BROWN AND GOLD BENEATH THE LIGHT.

EVERYBODY SCREAM. LET'S GO PADRE. PADRE.

PATRE.

Let's go Padre.

Let's go Padre.

PADRE.

Let's go Padre. I mean, it got the dubstep part right, but something fun for you to go play around with inside of Gemini. Right now, it's available in the

Gemini. Right now, it's available in the US and it's rolling out to YouTube creators in other countries. I don't

know exactly what that means or where YouTube creators use it yet. It does

appear that you can use it for free. It

says it's available in the Gemini app for 18 plus. And if you are on one of the paid plans like Google AI Plus, Pro and Ultra, you get higher limits. You

can generate more songs. I don't know what the limits are right now, but you can go play with it for free. But Google

wasn't done yet this week. They had even more. We also got this photo shoot in

more. We also got this photo shoot in Pomelli, which allows you to create studio quality marketing assets. So, it

says, "Pick a product. Start with any picture and don't worry about polish.

Choose a template. Select from

professionally curated templates.

Generate will automatically apply your business's aesthetics to generate professionallook images that feel on brand. And then you can refine, edit,

brand. And then you can refine, edit, and adjust your images with finishing touches. So you can try out photo shoot

touches. So you can try out photo shoot today. So they say, change the

today. So they say, change the background wall of this image. They pick

an image of somebody doing yoga. And

they say with this wallpaper, they pick a wallpaper. Click generate. And we get

a wallpaper. Click generate. And we get that person doing yoga in front of the wallpaper that they picked out. So if I head over to labs.goo,

google.com/pomelli.

I could click on let's get started. And

it wants me to enter my website. Sure.

futuretools.io. It's analyzing my website. It's going to take about 10

website. It's going to take about 10 minutes. And this didn't take a full 10

minutes. And this didn't take a full 10 minutes. It took maybe two, but it found

minutes. It took maybe two, but it found our business DNA. It knows the name of the site. It found the color scheme, the

the site. It found the color scheme, the font, the tagline, brand values, all of this stuff. Even found some images,

this stuff. Even found some images, although it's pulling images from tools.

So, let's go ahead and click looks good.

And we've got the new photo shoot feature here. Let's try the photo shoot

feature here. Let's try the photo shoot cuz that's the new thing they just added. Create a product photo shoot. So,

added. Create a product photo shoot. So,

we can select an image. I'll use this Glaciios image that I made in a past video of glass shard cereal. We'll go

ahead and upload this. Now, it's

choosing some photo shoot templates for us. So, let's go ahead and do contextual

us. So, let's go ahead and do contextual inuse ingredients and studio. Looks

good. And create photo shoot. All right.

So, this says it's going to take about 2 to 3 minutes. And again, it didn't take 2 to 3 minutes. I would say maybe 45 seconds. And we have some actual product

seconds. And we have some actual product shots of Glaciios. We've got one sitting on a table next to a bowl of cereal.

We've got somebody holding one over a bowl of cereal, one with some, you know, glass sitting around on the counter next to it. And then like a studio photo

to it. And then like a studio photo shot. So if you sell physical like

shot. So if you sell physical like e-commerce products, this pomelly is going to be a really great tool for you to go and use and toss your products into and let it generate these little templates for you. We also got a pretty

cool update to Notebook LM from Google this week. We can see here that the

this week. We can see here that the Notebook LLM X account posted this.

Prompt based revisions. Tweak, tailor,

and tune your slides just by prompting the revisions you want. So, if I head on over to my Notebook LM account here, let's go into our birds aren't real notebook here. I have a birds aren't

notebook here. I have a birds aren't real conspiracy slide deck that we can see over here. Let's click on revise.

And we can actually change slides by prompting the changes we want to see.

Let's say change the background to grid paper. will generate new deck. And we

paper. will generate new deck. And we

can see that it's generating a new deck.

It's going to take a minute or two here.

Cool. And a few minutes later, we have our slide deck. Let's take a peek and see if it did what we asked it to. All

right. So, look, it definitely changed this second page to a slide with grid paper. And all I had to do was prompt

paper. And all I had to do was prompt that into existence. So, that's a pretty handy feature if you're trying to build slide decks inside of Notebook LM by just feeding it a bunch of information, telling it to generate a slide deck for

you, and then you get in and you can then fine-tune individual slides to exactly the way you want them. I

definitely see that being a very helpful feature for a lot of people. And it's

interesting that uh Google didn't sort of make any big announcements outside of X about this on like a blog post cuz this seems like a really impressive feature, but I'm easily impressed. So,

who knows? And we got an even another model out of one of the big labs this week. XAI launched Grock 4.2.

week. XAI launched Grock 4.2.

Interestingly, they didn't make any sort of big announcements or blog posts around it. Elon tweeted about it, but

around it. Elon tweeted about it, but other than that, there wasn't really much fanfare around this one. But if I go over to grock.com here, I can change this from auto, and we can see we now

have Grock 420 beta. I can select this one, and it will use the new model. Now

what this model does and what makes this one a little bit different is that when you prompt it with a question, it essentially consults four different models, those four models come to a

consensus on the best answer and then it responds with the best answer from all of those four models. If I ask Gro a question like what's special about Grock 420, you can see it's got four different

agents here that are all thinking. So

according to Grock 4.20 20 itself. The

standout feature native four agent multi- aent collaboration. It's the

first Grock built from the ground up as a council of four four specialized agents working together in real time.

Grock, Harper, Benjamin, and Lucas.

There's a coordinator, a researcher, fact checker, real world knowledge. Uh

Benjamin is logic, math, code, and verification. And Lucas is creativity,

verification. And Lucas is creativity, hypothesis, and out of the box thinking.

They think in parallel, debate details, cross-check each other, and then reach a consensus before delivering the final polished response. Now, some benchmarks

polished response. Now, some benchmarks are shared here, but I'm actually not sure where these came from. They do site some sources, but all the sources seem to be just responses by other people on

X and nothing that actually officially came out of XAI or Gro, that's what we know about it right now. Leonardo AI, a company that I'm an adviser for, so I like to talk about them, got a makeover

this week as well. If you check out Leonardo.ai, it's a completely different branding and look. They've just

completely overhauled everything.

They're becoming pretty much the best all-in-one platform to do images and video and image editing and things like that. And it's just a much more modern,

that. And it's just a much more modern, clean, easy on the eyes user interface.

We did get some other large language models out of companies outside of the US this week, including Bite Dance Seed 2.0. This one has three separate models,

2.0. This one has three separate models, Seed 2.0 Pro, Seed 2.0 Light, and Seed 2.0 Mini. And these new seed models do

2.0 Mini. And these new seed models do claim to outperform on vision related tasks over a lot of the US-based foundation labs. They seem to win in

foundation labs. They seem to win in math vision, math kangaroo, which I'm not super familiar with. As well as VLMs are biased, and baby vision, not really benchmarks I know a ton about. Motion

and perception, they seem to beat out everybody else. Instruction following,

everybody else. Instruction following, they come close to beating out everybody else. And real world tasks, they are

else. And real world tasks, they are pretty much on par in a lot of them. not

on health and not on world travel necessarily, but on expert bench.

They've got a ton of additional benchmarks down here. If you saw my past videos about benchmarks, I don't always trust benchmarks because a lot of these models know how to sort of cheat and a

lot of these companies do know how to sort of train on the benchmarks, which I've gone over in the past, but I will link this up in the description if you want to take a deeper look at seed 2.0.

We also got a new openweight model out of Alibaba called Quinn 3.5397B- A17B. It's the first openweight model in

A17B. It's the first openweight model in the Quinn 3.5 series. It's natively

multimodal. And if we take a look at their benchmarks here, it's an openweight model that is apparently pretty on par with Claude Opus 4.5, GPT

5.2, and Gemini 3 Pro in a lot of areas.

Again, the thing that's most fascinating about this is how quickly open- source has essentially caught up to state-of-the-art models. I mean, the

state-of-the-art models. I mean, the state-of-the-art models hit a sort of level and then the open- source models sort of catch up to like right below them really quick. The state-of-the-art

moves further ahead, the open source catches up, and we just we're seeing this sort of pattern of these open- source models getting closer and closer to the closed source state-of-the-art models, which is really, really

impressive. All right, now let's move

impressive. All right, now let's move into one of my favorite pastimes, which is watching the reality show that is big tech company AI drama.

We can start with this one, but I won't go too deep into it cuz I did a full breakdown video on it. The Pentagon and Anthropic are sort of butting heads with each other right now because the Pentagon wants to be able to use

Anthropic's models for all legal use cases. And right now, mass surveillance

cases. And right now, mass surveillance and some use of autonomous weapons are technically legal, but Anthropic doesn't want the military to be able to use their tools for that kind of stuff, even

though they did sign a giant deal with Palunteer and the Pentagon. So, I don't know what Anthropic thought they were getting into when they signed up with Palunteer and the Pentagon. But

Anthropic doesn't want the US government to use their tools for mass surveillance or for fully autonomous weapons without a human in the loop. And the Pentagon is saying that's not your choice. We should

be able to use it for anything that's technically legal. Again, I did a whole

technically legal. Again, I did a whole breakdown video going into all of the details, all of the quotes, everything that's happening. So, make sure you

that's happening. So, make sure you check that video out if you haven't already. Last week, I showed off Seed

already. Last week, I showed off Seed Dance 2.0 know in my AI news video called this is the video model everybody's freaking out about or something like that where it's able to

generate very realistic videos of actual actors and various IP from these studios that are out there. Well, SAG After the Screen Actors Guild and whatever AFA

stands for, they put out a statement saying SAG After stands with the studios in condemning the blatant infringement enabled by Bite Dance's new AI video model Seed Dance 2.0. The infringement

includes the unauthorized use of our members voices and likeness. This is

unacceptable and undercuts the ability of human talent to earn a livelihood.

Seance 2.0 disregards law, ethics, industry standards, and basic principles of consent. Responsible AI development

of consent. Responsible AI development demands responsibility, and that is non-existent here. Disney also put out a

non-existent here. Disney also put out a statement saying, "Bite Dance's virtual smash and grab of Disney's IP is willful, pervasive, and totally unacceptable." The Motion Picture

unacceptable." The Motion Picture Association called on Bite Dance to immediately cease its infringing activity. They said in a single day, the

activity. They said in a single day, the Chinese AI service Seed Dance 2.0 has engaged in unauthorized use of US copyright works on a massive scale by launching a service that operates without meaningful safeguards against

infringement. Bite Dance is disregarding

infringement. Bite Dance is disregarding wellestablished copyright law that protects the rights of creators and underpins millions of American jobs. So

pretty much all of Hollywood, the studios, the Screen Actors Guild, the everybody is really, really mad at Seed Dance for even putting that out into the world. And apparently Bite Dance, who

world. And apparently Bite Dance, who owns Seed Dance, they also own Tik Tok and Cap Cut, is going to back off. Bite

Dance says it will add safeguards to Seed Dance 2.0 following Hollywood backlash. Bite Dance respects

backlash. Bite Dance respects intellectual property rights and we have heard the concerns regarding Seed Dance 2.0. We are taking steps to strengthen

2.0. We are taking steps to strengthen current safeguards as we work to prevent the unauthorized use of intellectual property and likeness by users. So, it

does sound like Bite Dance is not going to let people generate actual actors or characters from major studios and things like that. However, I did just get to

like that. However, I did just get to talking about how open- source stuff keeps catching up to the closed source stuff, and it's only a matter of time before somebody else creates an open- source model, push it out into the

world. People then download it onto

world. People then download it onto their computers and have access to it, and then there's nothing Hollywood can do. They can't get it removed off of

do. They can't get it removed off of potentially thousands or tens of thousands of computers. It will just be out into the world. So, I feel like Hollywood won this round, but the people that are developing a lot of these tools

and models are going to end up winning the war over time. Although, my hope is that there is some sort of like acceptable middle ground that the world agrees upon, right? If you look at the whole Napster thing that happened, if

you're old enough to remember that, everybody was pirating music, downloading it from Napster and Limewire and places like that, the music industry started suing like fans of the music and the companies that were making the music

available. there was all of these sort

available. there was all of these sort of legal headaches and it was a big deal and then you got Apple Music and Spotify and all of these places that made an easier alternative to actually pirating.

And I do feel like something like that will end up coming along where Hollywood and the movie studios sort of accept some level of letting people play with their IP while they still manage to get

paid for that IP. And it's something where it's easier than running open-source models locally on your own computer. So consumers will just end up

computer. So consumers will just end up using whatever that alternative is. Some

sort of Spotify, Apple Music kind of thing might be a middle ground that everybody lands on eventually. But I

don't know. That's kind of what I'm hoping for, but who knows how this all plays out. All right, I have a handful

plays out. All right, I have a handful more news stories I want to share, but these are all pretty quick ones that we'll run through in a rapid fire.

Starting with this news that came out over the weekend that the creator of OpenClaw went and joined Open AI. And

this one is really fascinating because this is like Anthropic totally fumbling the ball here. When Cladbot first came out before it was called OpenClaw, everybody was using Anthropic's Claude

models behind the scene with it.

Everybody thought that Claudebot was sort of a nod to Claude from Anthropic, but Anthropic basically sent them a cease and desist, made them rebrand from Claudebot. They became Moltbot and then

Claudebot. They became Moltbot and then eventually OpenClaw. And then Sam Alman

eventually OpenClaw. And then Sam Alman met with Peter Steinberger, the creator of OpenClaw, and ended up bringing him inhouse to help them develop the in-house agent stuff over at OpenAI.

This is another one that I did an entire breakdown of, which is why I put it in the rapid fire section instead of like the main portion of this video. So, if

you really want to hear all of the details, cuz this is a pretty fascinating and wild story, I did do a full breakdown video. The video is called The Cloudbot Story just took a wild turn. Definitely check that one out

wild turn. Definitely check that one out if you haven't already. It breaks it all down the whole timeline of things. The

company Manis that was recently purchased by Meta had a new update this week. This seems to be their response to

week. This seems to be their response to what is going on with Claudebot. So,

this says, "Introducing Manis in your chat, your personal agent everywhere you are." So, they're basically showing off

are." So, they're basically showing off here that you can use Telegram and directly inside of Telegram, you can have chats with Manis. And I think it also works with WhatsApp and whatever

messaging app you work with, you can go and tell Manis to do things on your behalf, which is the thing that everybody likes about Cladbot is that you can just go and chat with it and tell it to do things and it will go do

those things. So, this is clearly Meta

those things. So, this is clearly Meta and Manis going, "Oo, Claudebot was really, really successful. Let's do what Claudebot does now with Manis." This

company, Tavis, introduced a new like virtual twin kind of thing called Fenix 4. They claim it's the most advanced

4. They claim it's the most advanced real-time human rendering model ever.

Now, the whole idea behind this is that you can create these AI avatars that you can have conversations with through your computer, and they're actually sort of emotionally responsive. So, as you're

emotionally responsive. So, as you're saying stuff, they'll sort of nod and, you know, give facial expressions that make it seem like they understand what you're saying and what you're talking about. Me personally, I still think like

about. Me personally, I still think like something about the voices and the lip-s syncing just make it so uncanny and so like unhuman sounding to me, but we are getting closer. This is the worst it's

getting closer. This is the worst it's ever going to be. So, here's an example real quick. Notice how the guy who's

real quick. Notice how the guy who's talking to his little avatar here is like rubbing his eye. That's sort of important context about what he's going to say.

>> Dude, you look like you saw a ghost.

What happened?

>> Dude, I was walking to work just now and a bird came out of nowhere and pooped into my eye.

That's freaking disgusting. You need to go wash that out ASAP.

>> So, you can see that as he was saying stuff that he was kind of giving like a disgusted face and he noticed that he was rubbing his eye, so it was actually picking up what was going on in the other scene, which is what makes this

different and interesting. I just wish that lip-syncing voice over was a little more realistic. Like Seed Dance 2.0 kind

more realistic. Like Seed Dance 2.0 kind of nailed the lip syncing. I want to see that in avatars like this now, too. Now,

here's something pretty dystopian.

Meta patent AI that takes over a dead person's account to keep posting and chatting. Meta has been granted a patent

chatting. Meta has been granted a patent outlining an AI system capable of simulating a user's activity on social media, including continuing to post after their death. The patent describes how a large language model could

replicate a person's online behavior using their past data. According to the patent, the model may be used for simulating the user when the user is absent from the social networking system. I don't know why anybody would

system. I don't know why anybody would want this. This sounds dystopian and

want this. This sounds dystopian and weird and creepy and probably going to spin up all sorts of new mental health problems that we never expected it would

spin up. But, you know, social media and

spin up. But, you know, social media and meta have had a little bit of a tendency to do that in the past anyway. So, I

guess what do we expect? And finally,

I'm going to end on robots cuz robots are awesome. The company Unitary had the

are awesome. The company Unitary had the spring festival and they had these robots doing this like kung fu performance on stage and this is just so cool to watch. but also a little bit

creepy because, you know, this is probably the future of the military here if we're being honest. These robots are just mind-blowing what they can do now.

So, we can see them here doing flips off of this table, doing giant flips in the air. But what's really impressive, let

air. But what's really impressive, let me fast forward a little bit here.

There's a scene here in a second where they're all doing stuff in unison with like nunchucks and like, yeah, this is just absolutely wild. Again, I'll make sure all of the links mentioned in today's video are in the description

below. So, if you want to watch any of

below. So, if you want to watch any of these videos in their entirety, they will be linked up. But that's what I got for you today. A crazy week. Tons of new large language models. Again, mostly

marginal improvements for most people, but if you're a coder, a developer, well, you got some quality of life upgrades out of some of these new models. And uh it's been a crazy, fun,

models. And uh it's been a crazy, fun, exciting week. I don't expect it to slow

exciting week. I don't expect it to slow down. Kind of every week is like this,

down. Kind of every week is like this, which is why I never title my videos the craziest week in AI, cuz every week now is the craziest week in AI. But also,

I'm getting close to hitting a million subscribers. I'm like less than 100,000

subscribers. I'm like less than 100,000 away. So, if you want to help me get

away. So, if you want to help me get there, I'm probably going to do some sort of cool giveaway or like meetup event or something when I get to a million. So, if you want to help me get

million. So, if you want to help me get there and do something cool and be a part of the history of this channel, that would be super super awesome. So,

give this a like and a subscribe if you want to help me out there. Like, I would really appreciate it. But again, that's what I got for you today. So much fun talking about this stuff. I'm still

blown away every day that I actually get to do this for a living, watch and learn about cool AI stuff that's coming out and then turn around and tell you about it. Like, it's absolutely a blast to

it. Like, it's absolutely a blast to stay tapped in like this. Also,

sometimes a little bit scary and nerve-wracking and dystopian, but I'm still very, very happy that I get to do it for you. So, thank you so much for hanging out with me today, nerding out with me. I really, really appreciate

with me. I really, really appreciate you. Uh, don't forget to like and

you. Uh, don't forget to like and subscribe and click the notification and do all the things that all the other YouTubers ask you to do cuz it helps me out and make sure that you see more stuff like this in your feed. Thanks

again for tuning in. Really appreciate

you. See you in the next one. Bye.