KTibow

I think this is called tree shaking and Vite/Rollup do this by default these days. Of course, it's easy when you explicitly say what you're importing.

Does this replace ljharb stuff?

Right now Deepseek's official hosting is cheaper than everyone else who can manage to run the model, including Deepinfra. I haven't seen any good hypotheses as to why other than their large batch size and speculative decoding.

I've seen claims that the official API is less censored than the public weights.

Gemini 2 can purportedly do this, you can test it with the Spatial Understanding Starter App inside AI Studio. Only caveat is that it's not production ready yet.

There's a surprising number of extensions like this (see https://github.com/thomasj02/AiFilter, https://unbaited.danielpetho.com/) although it seems most are Chrome-exclusive

If I understand TFA correctly that's a claim it's covering and arguing against, not arguing for.

TFA doesn't use separate windows, only separate tabs.

I probably won't ever use this but the docs look really nice. Specifically: the header combines a blurry backdrop filter with something like `background: radial-gradient(transparent 1px, white 1px) center / 4px 4px` to make a background of transparent dots.

What's the HN etiquette around claiming that a reply is AI generated?

There are a lot of interesting options. Gemini 2 Flash isn't ready yet (the current limits are 10 RPM and 1500 RPD) but it could definitely work. An alternative might be using a fine tuned model - I've heard good things about OpenAI fine tuning with even a few examples.

Interesting eval but my first reaction is "using Mixtral as a judge doesn't sound like a good idea". Have you tested how different its results are from GPT-4 as a judge (on a small scale) or how stuff like style and order can affect its judgements?

Edit: they have a blog post https://pvq.app/posts/individual-voting-comparison although it could go deeper

Website authors can set a scrollbar gutter.

I personally predict this won't be as bad as it sounds since training on synthetic data usually goes well (see Phi)

I think that's an exaggeration, I know a few Google school districts and companies.

I've went through a few stints of using Bing. Eventually I end up starting all my queries with `@google`.

Not wdutch but Claude Sonnet is one of the best models out there for programming, o1 is sometimes better but costs more

Not the case, it doesn't ask even in a private window

Training and running fine tunes of Deepseek could get expensive.

Probably even worse, since I've heard that it's hard to steer away from the most common interpretation of a question.

The web tries to do a lot of things, even some of the stuff you named. There's a Bluetooth API in Chrome, server-sent notifications via service workers, and some attempts at VR support.

Funny that they're still using Claude 3 Sonnet then

Notably, solutions specialized for enwik9 (specifically fx2-cmix) take up only 110 MB, including the size of the decompressor.

I'm surprised nobody (in the comments I scrolled by) has brought up the Web option (which only shows links) yet

Most if not all of the times it "hallucinated" it was paraphrasing the web.

This has some meme potential, it succeeded at "show me the OpenAI data centers"

How long ago did you see it? I didn't start using it until recently and it seems it uses other placeholders today.

You actually can't pay for the latest models, they're only available as free with limits

Obligatory "uploading private health data to a random website is a bad idea" comment

When clicking or using "open in new tab" from the long press menu (not "open in new tab in group")?