I can't believe this is real

window.ai is the new API coming to Chrome for embedded Gemini Nano. I am impressed. There's a lot of potential here, but also a lot of silliness.
SOURCES
x.com/rauchg/status/180638577...
ai-sdk-chrome-ai.vercel.app/
Check out my Twitch, Twitter, Discord more at t3.gg
S/O Ph4se0n3 for the awesome edit 🙏

[@andrewzuo86]

Because Chrome didn't use enough memory before.

[@ivanfilhoz]

I wonder what the minimum requirements for a functional computer will look like in a couple of years.

[@gordonoboh833]

@@ivanfilhozthey would automatically be gaming laptops ready.

[@gordonoboh833]

@@coolemur976 computational resources, cpu clock cycles to process the input and output of the model, it would need to be loaded to a fast memory(RAM) for that computation to be fast enough that a user wouldn’t pull hair.

[@gordonoboh833]

Assuming it runs on cpu, most likely npu or gpu, either way you need RAM.

[@t3dotgg]

omfg

[@echtertill]

Alternative title: Chrome is getting more bloated

[@exceed.charge]

sequel title: edge copilot gone wrong! (gone sexual) (police showed up)

[@RoyerAdames]

lol but I like it

[@itamarsharify]

Alternative approach, browser is an OS

[@alexandreg.1000]

🚫 Add a compatibility level between AI software and browser ✅Add AI software into browser ...

[@RepChris]

@@itamarsharify EMACS is quite the decent OS, its a nice webbrowser too, just lacks a good text editor out of the box

[@tinrab]

The use of the word "delve" in comments is about to skyrocket.

[@Interpause]

lets delve into that phenomenom shall we?

[@le9038]

?

[@ajmaln]

Lets embark into delve

[@CaptainHottub]

@@le9038 ChatGPT tends to say "delve" a lot more then humans do. Apparently, in 2022, the number of scientific papers published with "delve" in the abstract or title was 0.056% then it shot up to 0.789% in 2024. Meaning, it is very likely that ChatGPT or another LLM was used to write some of them.

[@cherubin7th]

Delve into delve delve

[@Cyber-Talks]

It's not meant to be used for this kind of questions I think Theo missed what's written next to the feature flag he turned on: "This API is primarily intended for natural language processing tasks such as summarizing, classifying, or rephrasing text. It is NOT suitable for use cases that require factual accuracy (e.g. answering knowledge questions)."

[@malikau917]

Thanks for pointing out, I’ve missed this as well

[@ZoranRavicTech]

Those are still a pretty challenging tasks. I doubt you can achieve a reliably good result with a small model. You're still better off doing it on the server or distributing a desktop app, so you can make full use of the gpu.

[@Cyber-Talks]

@@ZoranRavicTech Distributing desktop app is not always an option. While doing it on the server can be a better solution in terms of the quality of the answers this has a computational cost which is not small in lots of cases. Maybe doing it on the server as a solution for those not using Chrome is a valid overall solution. How other browsers can support this kind of feature it's also a question I'm sure we have a good answer yet( how it can become part of the web standard?)

[@ymi_yugy3133]

I don't like this at all, because it feels pretty fundamentally at odds with the idea of an open web. How is the W3C supposed to come up with a reasonable specification for an AI model that can be implemented independently by multiple vendors? How are developers supposed to build websites that offer a consistent experience across browsers? What about backwards compatibility. The web tries really hard to not break content. How can browsers make sure that an update to their model doesn't break any websites? Even more fundamentally these small embedded AI models aren't very general, but have to be highly tuned for specific scenarios. Browsers obviously can't ship models for specific scenarios. So here is what we get. A shitty, browser specific language model, that breaks with every update. If I had to guess the only thing we get out of this will be small demos and companies running their crappy customer support chat bots on the client to save on server costs.

[@ea_naseer]

The specification would be led by Google most likely

[@ymi_yugy3133]

@@ea_naseer I don't mean the specification in terms of the API, but it terms of capabilities. If I build a feature on top of this AI model the same prompt should give me similar results regardless of whether I'm doing it in Chrome, Firefox or Safari. That's the part that is really difficult to specify.

[@dalehorton7748]

This is kinda why Chrome exists in the first place, Web standards were being written, browsers like Internet Explorer simply didn't care, and browsers like Firefox cared more about retaining compatibility. Google cracked the shits and wrote their own standards-following browser as a tech demo, and it became the only "proper" implementation and we all adopted it. Alternatives took many years to catch back up. The web adopts whatever works and gets used, it can't get used until it exists, and Chrome has often led this charge.

[@averagegeek3957]

@@dalehorton7748 And now they are trying to become the new Internet Explorer by adding non-standard behavior that will break sites on every other browser

[@itsmeagain1415]

@@ymi_yugy3133 by the fundamental nature of ai this is impossible! so you are effectively dismissing indeterministing computing as a whole from having the convenience of a specification

[@eUploads]

this is gonna be weird for cross browser support

[@Kannatron]

Could allow handoff to an os based model in the future I hope

[@Jaryt]

@@Kannatron Good thought, but even that'd be weird. The main issue that if you're using let's say a windows model, vs an apple ai model, you'll get very inconsistent results. (same if you have a firefox embedded ai vs chrome embedded ai) ...almost like it'd be a need for a new browser standard that no one would ever agree on lol

[@cerulity32k]

god i hope this doesnt get to firefox

[@husanaaulia4717]

Why though? Can them just follow what Chrome did?

[@thatanimeweirdo]

​@@cerulity32kThey are sadly very fond of AI lately

[@JohnDoe-jh5yr]

What could possibly go wrong?

[@JohnDoe4321]

I asked the AI the same question. and it said nothing could go wrong. So we're OK 😁

[@inononeeee]

access our saved passwords and post promotions on our personal accounts, we don't even know who?

[@cerulity32k]

-r /

[@ffuyu]

DISASTER STRIKES

[@gamefun2525]

the dependency and trust on the knowledge of the model. it can blurt out random inaccurate shit if not used correctly

[@bill_and_amanda]

We did it y'all. We found the one way to make Chrome use even more RAM.

[@LoopikCZE]

But who cares I hame 40GB on my desktop. Ram became so cheap lately, that it doesn't even matter. It will force Apple and others to sell ram for less, because 8GB is joke today.

[@sansmoraxz]

@@LoopikCZE my workplace cares where we are forced to use systems with 8 gigs of ram.

[@leeroyjenkins0]

​@@LoopikCZEhave you heard of smartphones?

[@ahdog8]

I feel like Chrome is just going to keep adding stuff to the API that only their browser supports so the other browsers fall behind and they get 100% market dominance

[@alekkrstic]

I feel like the inverse would happen if they start deviating from standard

[@AndrewTSq]

Are there apis today that only Chrome can use?

[@Reashu]

​@@alekkrsticexcept that they run enough of the web to just "break" other browsers

[@pelic9608]

Microsoft tried with IE and failed. Google will with Chrome, too.

[@zBrain0]

​@@AndrewTSqI daily drive Firefox, and it is fine 99.99% of the time. But there are definitely websites out there that don't work quite right. It's hard to say whether that is because of some deficiency in Firefox or if they are doing something in a chrome specific way. I can say that most web developers nowadays probably only test on Chrome. Which would lead to Chrome basically being the standard. If something breaks in Firefox it is more likely that Firefox will have to change the behavior of their browser to match Chrome rather than web developers making sure their stuff doesn't break. All of that to say, if Chrome introduces a feature like this Firefox will need to have some sort of approximate equivalent because every moron web developer is going to jump on this train and nobody's website is going to work anymore without it.

[@Alcaline-hu2vu]

Please don't let this become industry standard, fuck that, I do not want AI models running directly on my machine, specially because I know the React devs are going to go absolutely nuts with this and make websites 10x heavier. Also people using older devices (me) be damned I also feel like this is breeding ground for a new fragmented internet era like in the early 2000s, when every major browser had its own specific APIs so most websites would only work on one browser, but then there's like 3 of them that only work on the other browser, so now you either can't access half the internet or need 3 different browsers on the same machine

[@DoubleRainbowXT]

dont worry you can always disable it in flags also chrome cant add it on mobile if you dont habe a lot of ram.

[@test-rj2vl]

People might be tempted to leave AI prompts in code - because who would like to study 20 different JS frameworks that all have high learning curve just to get Hello World done in NodeJS. Meaning that then all the browsers have to support AI or some sites are just broken.....

[@achillesa5894]

Would you rather have all the AI running on a random server somewhere which doesn't work while offline, requires high bandwidth, and might decide to go down for no reason? Local is 1000 times better. Now do I trust google with it? That's a different question. This is a great idea but it should be an open standard.

[@ZipplyZane]

@@achillesa5894 It should be quite clear that we'd rather have websites not use AI at all, unless it is absolutely necessary. The only way I like this is of websites have to get AI permissions, and that they get to specify the AI model.

[@malikau917]

You are too old, close web page and go read paper books

[@DanielMelogpi]

I detest the idea of a Google model underlying the implementation of a global method. What about other browsers now and in the future? How about the spec of what to expect from this? It nevers works well when vendors add they proprietary shit like this

[@calimio6]

back to 1999 internet

[@frenchButt44]

Why not, don’t use if you don’t want or use firefox or some shit..why cry.

[@kenwarner]

They should make the model configurable like default search engine setting or something

[@vaisakhkm783]

I am sure it user will be able to download multiple models... firefox also doing a similar thing, but for image alt text..

[@skeleton_craftGaming]

I love how Google learned nothing from Microsoft doing this exact same s*** with IE in the early 2000s..,

[@IamProcrastinatingRightNow]

We already reached a point in which there are only three browser engines left. Chromium, WebKit, and Gecko. But if different browsers include different AI models that lead to different results, it might be very quickly hardly possible to use anything but Chrome. Assuming even the other Chromium engine browser, like Chromium or Edge, do not have access to the same model. And when there is only one browser left, the war on ad blocks is won by Google. And so much more. User tracking simplified for Google and Google alone. Maybe predictive modelling? That's the right moment for an "outstanding move" meme.

[@opposite342]

Technically there's also Servo, but it's very early and ambitious in its project. Regarding browsers, Google has been and is going to continue winning. So not much others can do either way imo

[@test-rj2vl]

And now combine it with Google's web DRM. We need some sort of antitrust lawsuit to make Google sell chrome and prevent them creating a new browser in the future.

[@MatthewSwabey]

Yeah the whole trusted browser DRM where it verifies the operating system and extensions and everything before rendering the web page is kind of terrifying

[@echtertill]

Yeah it sounds cool if you're on the AI hype train but in the end it's just more bloat, especially if you want a lightweight / minimal setup...

[@avhiro]

my thoughts exactly

[@RepChris]

If you want a lightweight/minimal setup you dont want chrome anyways so ignoring the fact that everyone is going to want to mirror chromes features there isnt any loss

[@consciouscode8150]

I'm on part of the hype train and even I think this is like 4 years too early and contributes to bloat/global namespace pollution

[@tablettablete186]

​@@RepChrisWhat is the lightweight alternative?

[@victorbitencourt9481]

Maybe a terminal web browser like Lynx 😂

[@trietang2304]

Just hope they don't actively train that model when we using our browser.

[@akam9919]

...you already know the answer to that.

[@vaisakhkm783]

They can't, our gpu sucks for that it's much easier to sent data to server and training it, which they already doing

[@xaf15001]

​​@@vaisakhkm783They can't, but now they have more reasons. "Hey bro, you used the AI a lot lately. Can we get some data as described in [comically long lawyerspeak]?"

[@vaisakhkm783]

@@xaf15001 i read entire terms and conditions of meta ai when it came to whatsapp, where it's written, if we accept to use ai, our private messages will be shared with 3rd party vendors, and might be reviewed by humans.., so ya

[@tschorsch]

​@@vaisakhkm783 I would assume that that is what the OP meant.

[@sleepykitten2168]

It got a lot of info wrong. For example, the new england patriots did not win the superbowl in 2001. The baltimore ravens did.

[@MrEnder0001]

1. Its beta things will change and improve there is a reason its not even on canary yet, its for devs only rn to learn about 2. It is all instantly ran locally on his computer so it may not be used for information but maybe for generating content like recommended tags for a video given the title or something

[@null_spacex]

Oh my, a feature not ready to be used isn’t ready to be used. Who could have imagined that?

[@akam9919]

​@@MrEnder0001 i know this is nitpicking but, excluding the fact that an offline ai model without access to a database of modern relevant topics and associated tags would likely give poor tag choices, recommending tags based on the video title would not work for like half of youtube clickbait.

[@outtrigger]

who cares who won the superbowl. as long as it gets the important stuff right thats all that matters

[@t3dotgg]

The model really likes the Patriots

[@andrew_ray]

I will submit that if your business model involves spending more on parts than you can charge for the product, you have a BAD BUSINESS MODEL.

[@rollthedyc3]

This is terrible. It's not standardized. It's just chrome imposing it's dominance again.

[@stepans2167]

It's literally behind a flag, what are you talking about

[@mtx33]

@@stepans2167 for now, as usual

[@MrHamsterbacke756]

Yes, because feature flags are meant to stay forever.

[@meru_lpz]

"How terrible!! Chrome being more feature-rich than the competition!!"

[@stepans2167]

@@meru_lpz to be fair, standards are there for a reason

[@pencilcheck]

This will go very wrong and misuse in the future lol

[@tnypxl]

I just need my browser to render web pages and make keeping track of the ones I like a pleasant experience. I don't need it to do any more things than that.

[@raymond_luxury_yacht]

And what happens when you don't need web pages anymore because the model will render everything in a format that's best for you, not the designer? The web is dead. Real time rendering of content retrieved from database and constructed on the fly is what the future holds. Thank god. Imagine the entire website industry gone.

[@rasalas91]

autocomplete, TL;DRs, translation, ask a webpage/pdf, ...

[@Aoredon]

@@raymond_luxury_yacht Keep dreamin' buddy

[@tnypxl]

@@raymond_luxury_yacht Thanks for the laughs.

[@tnypxl]

@@rasalas91 Literally what extensions are meant to provide. This does not need to be part of the core browser.

[@Cool_Goose]

I wonder how the heck one would do debugging between different browser versions and what I assume might be different llm implementations.

[@thedreamingtechie]

Yeah, it would be super difficult ensuring you get consistent outputs across different browser implementation (probably bundled with different LLMs). Prompt engineers would be excited about this 😂

[@LaughingRam]

My browser ai told me to set my house on fire.

[@JohnDoe4321]

You asked how to get rid of termites. Setting the house on fire got rid of the termites. AI answer was 100% correct. 😁

[@LaughingRam]

@@JohnDoe4321 the prompt was actually "my wife sleeps a lot. What's the best way to wake her up?"

[@COArSe_D1RTxxx]

​@@LaughingRam Well, doesn't that do the trick as well?

[@xgui4-studios]

@@LaughingRam lol

[@vaisakhkm783]

Before ai, a guy on reddit too told me the same... so... not much different...

[@joszze]

Nobody can ignore this thumbnail

[@renealbrechtsen9743]

I don't want this built into my browser.

[@Aoredon]

cry

[@AGentooUser]

Even if the feature is good, All of us became too fed up with ai

[@freedustin]

You: "Chrome" Me: *Tab Closed*

[@naturegoggle]

They should add google translate built-in in the browser first.

[@Reashu]

Isn't it there already?

[@naturegoggle]

@@Reashu yes. I meant the one where user can type and translate. not just websites translations.

[@atxorsatti]

Imagine a world where not only you can have a memory overflow in your browser, allowing for rce, but also a context flow because all thr websites talk to the same model. Yay.

[@Farstors]

Today in another episode of... "Google fucks with web standards"

[@lpls]

Chrome is the new IE

[@dariofagotto4047]

Context? Who needs it; I really hope this will download the model (or hopefully models) on demand and not on install (with maybe hints in manifest for offline webapps using it), I don't want like every electron app to become 1GB larger just cause

[@jordanmancini]

It's truly amazing how it got all this information about images that weren't there. Almost like it was hallucinating

[@_evillevi]

Concerned about the idea of nano and it's training data not being open enough, and find the idea of a proprietary model being part of a web standard. Will _google_ it later

[@AvanaVana]

11:40 you mean “the background is dark and _undefined_ “ 😂 Also, this video is giving me vibes of way back when I downloaded and built from source an early chatterbot written in C (can’t remember its name…it was in the vein of A.L.I.C.E. and such)…this would have been 1997 or 1998.

[@J-wm4ss]

ELIZA? It was supposed to be a therapist and was very primitive

[@adriablancafort]

What a time to be alive.

[@jazzymichael]

Awesome overview!

[@aeronwolfe7072]

you have a mounted drive called 'Incel'.... lolol 0:59

[@AntonioBrandao]

lmao

[@comosaycomosah]

lmao good God

[@woet7891]

Having it work across browsers and standardized will probably be a battle of many years for Google and others, but loving the idea of having access to a cheap API like this in my apps.

[@thomassynths]

might have been useful in a tool like node. I think its worthless in a browser. As a user of a website, I would disable the beature. If I want to run a local llm, I dont need chrome to do that.

[@akam9919]

I can see one case where offline AI stuff could be useful...but if you can't mix and match whatever AI model you want in there... but it should be controlled by the user and not the browser... That being said, most AI features are total BS or near useless. There are really just a few dozen of use cases at best...but they are not necessarily something an LLM would be good at dealing with. If I want AI, I'm either running my own models locally in a non browser environment because I have horse power and have the time and energy to give a shit about it for five minutes, or I'm using a remote one because I don't have a good enough model or the compute power. Therefore, this thing is ultimately useless.

[@user-ti8hb1if6i]

Yeah, this is really cool, really. Now, recall there is a thing called W3C. How are they supposed to come with this API's standardization? Or should we let each browser implement their own and write several app versions like people used to at the time of IE/Netscape war?

[@MatthewSwabey]

I wish W3C had not tried to replace the whole internet with an XML standard, forcing the main browser companies to just create HTML5 instead. I think it really weakened their authority and has led to a lot of fragmentation.

[@SanyaZol]

The background is dark and `undefined`... Gemini seem to be learning from javascript

[@flamakespark]

Oh boy, we'll have a lot of fun creating polyfills for this one...

[@HighonLife-gh7gj]

dude roasted the gemini models at the end lol

[@IlyesCodes]

My boy got me enlightened this morning

[@brianressler569]

Could be good if it was some sort of standard across browsers with wiggle room for different models, definitely could make it vastly easier to setup some sort of chat app which I've had an interest in doing as a fun side project. Having the ability to turn it off would be good for privacy if it does become a standard though, I would hope that wouldn't be buried in the dev menus.

[@nikkehtine]

Good to see Chrome entering its Internet Explorer arc

[@s1v7]

this is literally the future of any app

[@Fanaro]

These dynamics will also give Apple such an edge in the AI wars...

[@Vi51576]

This feels like it might turn into background crypto mining 2.0

[@OlafsLeftArm]

And we have just found a way to make every app 10Gb when everyone want to ship their own ai model 😂

[@LeahLundqvist]

cant wait to delve into this!

[@theisoj]

this is really unreal!

[@schtauffen5975]

i think that ai allows for cool things but will be a net detractor from the human experience

[@andrere__]

this is awesome

[@alexholker1309]

A model-agnostic API to ask your browser's AI something is a nice idea. I'd personally just want a dummy AI that returns null to everything, but you could have extensions for various embedded or outsourced AIs.

[@mian1222]

and i cant believe youre in my home page after pressing not interested 4 times but life sure has its suprises

[@Thelukkest]

Pretty great

[@torsten_dev]

Yeah building a web standard around closed source AI models sounds like a great idea. That won't enshittify the web _at all_.

[@PrajwalDSouza]

This is awesome. Function calling for free finally!

[@deliriantabuser]

very cool!

[@user-kp4sf9lc2u]

It's good for some sorts of in browser RAG things

[@Karurosagu]

The picture is undefined

[@almcchesney]

lol younger devs are so funny. "So what if we could install to one cached location?", Bro just described M$'s GAC from back in the day.

[@Yours--Truly]

This is the birth of vulnerabilities

[@m12652]

Spooky... but also very cool. How long before the user can create their own automations I wonder.

[@ScibbieGames]

Finally, we can store the virtual DOM in there.

[@b.o.t7888]

How impressive

[@isan-sunshine]

that model is high

[@riendlyf]

Wow, bright future for ui devs. Seems like ai will generate html and css on the fly

[@bobwilkinsonguitar6142]

When Chrome finally integrates it, the users could switch AI models in default settings, like we do with search engines, and devs would never have to worry about integrating specific AI's

[@thygrrr]

"The median website in 2025 downloads 539 kiB of javascript and 2.18 GiB of language models."

[@AffectionateCanoe-vc4hj]

Alternative title: g4f is getting more providers

[@rverm1000]

Ai built into google colab has been useful on some coding problems.

[@erb34]

Now you can hallucinate offline. Would be great for firefox to create an api and way to use llms too but you can select the model. (qwen/llama, etc)

[@tejashp98]

very cool :)

[@pelic9608]

Time to mention Lynx again.

[@alexanderbr8525]

I think there should be a way to use the api with a custom model. But i dont have a good solution to caching, which would be needed

[@ottergauze]

Can't wait until all the Chromium Electron apps I have installed suddenly grow by 400MB.

[@Gentoo701]

Bro, can't wait until all electron apps will ship their models.

[@Spinikar]

This is cool, but at the same time, I wonder how practical this will be. LLMs can be pretty unpredictable. Still want to play with this, though. I'm wondering if this could make language translation of site easier. Just maintain your site in one language and let the browser built in AI model translate. I know there are tools for it, but I find LLMs are a bit more reliable for translation.

[@cherubin7th]

I would like that I can have one shared model on my OS level that can be used by all my apps.

[@heckerhecker8246]

setInterval looking... very nice right now-

[@markuszeller_official]

Oh no! No more Lemmings.

[@neoqueto]

I simultaneously really would love having something like that in a browser and I REALLY FUCKING HATE THE IDEA OF THIS

[@creaky2436]

Chrome is by far the best browser. You’re crazy for that. Thanks for the video though! Interesting as always.

[@NuncNuncNuncNunc]

Who won the Superbowl in 2001 - The Baltimore Ravens. It was Ravens vs NY Giants, 34 - 7. Once again proving the point that LLM are only useful if you already know the answer.

[@larsfaye292]

Just remember: EVERYTHING an LLM outputs is technically a hallucination. Just some hallucinations are correct, and others are not.

[@abdallahmeddah8461]

I think this feature is very cool, I know it's gonna eat up a lot of memory but eventually a solution will be found for it. I see that It makes possible what was not possible. It enables processing data that wasn't processable before LLMs, what I mean by that is that prior to LLM, if you don't know the exact structure of the data you are going to get as an input AND all the possible values that you might get as an input. You can't process the data. Now, we can, for example say I want to check if a page I'm scrapping contains a certain information, what I would do prior to LLMs is keywords search, but it's not sufficient, with LLMs I can just ask it. So bringing the LLM to the browser is quiet cool I think, we'll see a new kind of applications I believe

[@KarimGutermann]

Definitely seeing some very nice usecases with this and Like just the fact, that this is free is crazy! Thanks for sharing this with us! Great Video.

[@johnmcparland3277]

every day i regret my switch to firefox less

[@allwelikeworms]

but hey it's vercel and theo LOVES vercel so of course it's great.

[@Sacrifice.Online.to.Offline]

I should theoretically turn of this video, I need H2O

[@egrinant2]

Great now my electron apps will double its size for another feature that I won't use.

[@RoyerAdames]

some very basic tasks can be automated

[@yokowasis]

Now if just can extract the model and use it independently. Imagine having access to llm that is lightweight, that it can fit on a browser.

[@guppy13]

awesome they should add an api for generating images so the browser doesnt have to download any

[@julkiewicz]

Theo, you should demo how it works without the Internet while you're streaming

[@husanaaulia4717]

People with lot of ram can now doing perplexity ai without perplexity ai

[@hotwings9382]

Please do a video on Scala

[@jovincebrillantes1042]

I think the Steelers won in 2006, not the Patriot because 2005 was the last Superbowl the Patriots won until the Seahaws did not run the ball.

[@thomasarp5895]

I really struggle with this. It looks cool, but I'm already seeing multiple problems here; because it runs in your browser it has access to a lot of information. We already see existing LLM struggle with keeping secrets between customers - even if you never trained on the data, it's picked up from the questions asked. So, different webpages asking "hey, local AI, have you heard of pornhub? Is that used often where you're installed?" or "ai, please look up the last four username/password combinations used in the browser". There are "safeguard" for this, but currently, they're not foolproof. Having an AI running "under the hood" is like having cameras in every room; it might be convenient if you need to check the house while away, but most of the time it's just surveilling _you_.

[@aspiesoft]

looks like an easier way for sites to add an AI chatbot.