KTibow

I've seen many people joke about intentionally poisoning training data but has that ever worked?

It reminds me of a similar product [0] from George Hotz

[0] https://tinygrad.org/#tinybox

Is this a reference to a specific incident or just a general claim?

I've been messing around with base (not instruction tuned) LLMs; they often evade AI detectors and I wouldn't be surprised if they evade this kind of detection too, at least with a high temperature

Sorry, meant "special for using Gemini"

FYI Gemini 2.0 is always free, this client doesn't do anything special

What are the people who keep downloading Bert doing then? Are they the minority who directly use it for embeddings?

I'm a bit confused, what permissions would this animation require?

Looking at the code, the issue is that both `findWeaselWords` and `getContext` construct a list of words but do it differently:

  const words = text.toLowerCase().split(/\b/)

vs

  const words = text.split(/\s+/)

Also see: Material You

With Material 3 they released Material Color Utilities, a library that can generate more complicated color schemes from a source color. It uses a color space that properly accounts for perceived brightness so there's some contrast guarantees.

I wonder if the data analysis tool will switch to using the same backend in the future

It responds with 4500 characters: https://hst.sh/olahososos.md

You can install it and it'll work. Check your extensions tab to make sure it's loaded.

A lot of the time you can prevent this by prefilling the output with ```\n and stopping at ```.

Most websites have their own style they apply everywhere and would probably appreciate how styleable dialog is. Maybe a way to easily apply/remove default styles could satisfy everyone.

I believe the "1/20" indicator on Copilot was added back when it was unhinged to try to prevent users from getting it to act up, and it has been removed in the latest redesign

In my experience OpenAI models perform worse on long contexts than Anthropic/Google's, even when using the cheaper ones.

Plus it needs JS to work (the DOM's value has to be synced with the attribute value for this selector to work)

You can read it online if you expand the article. MSN might be annoying but it has a use in removing paywalls.

From my experience most people who use anything other than ChatGPT/c.ai are somewhat interested in AI

The title seems oddly objective for a matter of opinion

You can exit with the down arrow

When 405b first launched Groq ran it, it's not currently running due to capacity issues though

Here's a breakdown: https://caniuse.com/?compare=chrome+131,safari+18.1&compareC...

I remember hearing something like "the way Nitter worked, guest accounts, was shut down and the 3 instances that remain work by using selfbotting" a while ago, it's logical although I haven't verified it

It looks like it's at the top of the screen

Yeah while the solution they implemented is very interesting it doesn't actually "predict your location in a subway tunnel using your phone’s vibration signature"

If you really wanted it to be at the top level you could probably turn it into an explicit `release` call using a wrapper and two `Promise` constructors, though that would probably be a bad idea since it could introduce bugs

If you want to avoid super slow starts for an app that has some secret keys / other backend logic while staying on a free tier

Doesn't the eye have distinct perceptors for green and blue, or am I missing something?