I can't believe this is real

Because Chrome didn't use enough memory before.

Alternative title: Chrome is getting more bloated

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

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)."

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.

this is gonna be weird for cross browser support

What could possibly go wrong?

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

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

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

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

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.

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...

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

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

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.

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

This will go very wrong and misuse in the future lol

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.

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