3836293648

It's literally a 3.0.0. Why have a major version at all in that case?

American copyright, as opposed to most European copyrights, only protects creative works. Fully machine generated weights are not creative as no human touched them. You can copyright the training code, you can't copyright the output.

There is an argument that it's a derivative work and it's therefore copyrighted under the same protection as the training data, but that gives the rights to the owners of the training data, not the ai-lab. Courts are seemingly mostly rejecting this though as it's different enough to be fair use.

1 page = 0 clicks, just scrolling. What you mean is called "above the fold"

And their names are so frustrating. We don't run 16 bit systems any more. Word is a word that means something and I don't care what a word was on the 8086. The only honest things the WORD macro could expand to are size_t and ssize_t. 64bit is not a quad word, it's just a word.

And in some contexts you have the subtle distinction that fold is linear and reduce requires an associative operation and an identity element (aka a monoid)

Yes! This is by far the number one program I've missed since I left Windows years ago. GIMP is so unfriendly, krita is for drawing, pinta is a very shallow clone that only just stopped crashing a few months ago. Paint.net is just better in every way.

~1960

ELIZA beat the Turing test and then everyone forgot about it. Humans are just really terrible at recognising robots.

Intelligence is still not well defined, and by more solid attempt LLMs absolutely are.

Intelligence, Knowledge and Sentience are three very separate concepts.

They moved from HBM to dedicated silicon and only got a 48x speed up? That is so, so, so much less than I would've expected. Any numbers on how it scales?

There's a zed minus the iffy bits fork called gram. Haven't used it though, I'm still stuck in Emacs trying to figure out how to get out.

Anthropic basically downgraded all their tiers when they released Mythos, no nah, Sonnet 5 is the successor to Haiku 4.x

Emergent phenomena = pass/fail grading of multi part problems. If you're just gonna count it as a fail you're not gonna see the progress until everything works, even if every single subproblem improves linearly.

No? That would mean implementation defined, not undefined.

No, you can't assume it. You can trust them as they made that claim outright. Or you can choose to not trust them, I guess

It's not potentially, it's settled. At least for now as neither case wanted to move on to appeals

Oh, the context window I overrode immediately, the default one can't even reach my prompt in opencode, it's pushed out by all the pre- and post-amble. But I should probably try something less compressed next time I bother.

Or, as others have already noted, it's only about 15k and the repo includes tools and test programs.

An ASIC won't help until models are compute bound. They're still memory bound at all levels.

Yeah, I really should know. But I'm using whatever ollama gave me by default, which probably isn't optimal.

Slow means CPU only in this context.

Maybe not afford new, but they probably already had it from before the current crisis?

Could you expand more on what you do with qwen3.6? Because I couldn't get the denser 27B version to do trivial "take this pattern, repeat it over a single file with minimal thought, just slightly beyond what I can do with sed" reliably.

I am not complaining about it being 17MB. I am complaining about someone considering it tiny.

> Tiny. A single ~17 MB app

Oh goodness what have we come to? I know we're comparing to electron monstrosities, but still

I don't think anyone complained about the language changing, it was a lack of trust in the language due to the compiler not getting any updates. You can still improve the compiler without having to change the language.

No, that's just a convention. Any function can just create a new allocator, or import the global one, or use malloc from libc. Zig does nothing to enforce this convention.

That's just not true. It's just a convention. You can assign the allocator to a global or use malloc from libc. The article is correct.

I think you just happened to miss it. It's very commonly mentioned in the new systems space, alongside Jonathan Blow's jai.

So per subagent*.

Vad himla pratar du om? ALLT använder dd/mm-yy. Jag har aldrig sett något på svenska som använder yyyy-mm-dd, det finns bara inte. Det finns i Sverige, men då är det också på engelska.