ironqcold

What a cool project

Interesting. I didn't think there was a link between cancer and infections.

Wow. That was interesting.

I don't quite understand what this is about

That was interesting

I'd want to see p99 at 1k+ QPS on the same scale

I wouldn't say we've settled on a specific set of languages

I didn't know about this, that's interesting

6,500 instructions is a small training set, so I'd want to see how it performs on unseen indicators

Funny that the 2B loses to the 0.8B. Question about the benchmarks: BBH and JudgeBench are more reasoning, where you fall behind, but for zero-shot classification there are more relevant ones like Banking77 or CLINC150. Was there no temptation to pick something closer to where System 1 models are actually used?

Completely unusable for that, yeah, that's a 100M model, not an assistant

How much of the tools you use every day stays unnoticed...

Fair distinction

The critique of CLI-oriented SDKs is fair. I'd want to see how it behaves on long-horizon tasks where the "no sub-agents" constraint starts to bite

The thing I'd actually want to know: is Opus 5.5 Medium genuinely equivalent to Astra High on real work...

Modern AI is still far from what could be called intelligence. But I think that one day this, too, will be created; the only question is when

I’m actually curious now. Thanks, you’ve given me something to do this evening.

Google is increasingly locking down Android. I wouldn't be surprised if they killed off AOSP entirely in a few years...

My takeaway: we're degrading the web for real people to slow down bots that will just move forward. The solution seems is worse than the problem. At some point, we need to accept that the open web as we knew it is dying...

Interesting project. If it actually works it opens up a ton of possibilities for testing and reverse engineering.

Oh, well... It was bound to happen eventually. I suppose there must be some advantages to such a solution.

Thanks for the collection. Racter was especially fun — his texts sound like they were written by a bot that had too much to drink and is trying to sound deep. But back in 1985, that was probably pretty impressive

Cool plugin. I got annoyed myself when Kindle cut off long highlights. The fact that the author managed to pull them out using the Mac's local database and OCR is solid. And good on them for not using third-party services for recognition, but going with Apple Vision instead — fast and no extra hassle. Shame it's tied to macOS, but for Mac owners it's perfect. Anyway, useful thing for those who actually read a lot and take notes.

Looks like an atmospheric explorer kinda thing. Low-poly SF where you can just wander around and take in the city, pretty chill. I dunno, I really enjoyed it

I compared. On the color version, the text seems to have a slightly fine mesh or rippled texture. There's no huge difference, but I prefer the black-and-white version

Well, this is a rare case where a grand juries declining an indictment. They usually churn out indictments like a conveyor belt but here they refused. I wonder why?

The scale of problem seems a bit overblown. Anna's Archive paint a picture like AI companies are some movie villains burning books so no one can see them. but in reality they just disassemble them into pages because it's cheaper and faster to scan. Most of these books is highly specialized, they been collecting dust on shelves for decades and nobody need them.

But the problem is real. Even if these books aren't needed by anyone right now, them digitization in a single copy that end up behind seven locks at a corporation is not great, because AI doesn't replace the original. You can't to ask a neural net to give you a exact copy of a page from that book. So yeah, the post dramatizes a bit, but the point are valid. We need open digital archives.

I really can't disagree. What's the point of all this talk if for most ordinary people these technologies do more harm than good?

In general, feeding the AI what you write for verification sounds like a good idea. But I think even this should only be done by someone who understands exactly what they're doing, how, and why, because otherwise they won't be able to analyze the accuracy of the AI's responses...

I think we're still in the early days of figuring out how to use AI as a daily work tool. It's going to take time to develop the right practices. I do see the point in having token limits, they force you to be intentional. My guess is that eventually we'll end up with small, local LLMs that are unlimited for routine stuff, and we'll save the big expensive models (with quotas) for the heavy lifting.