anematode

5 ms·

1.
by anematode 13d ago
That makes very little sense when you can just use WebGPU and have fine-grained control... for something like LLM inference, having this kind of abstraction seems like quite a hindrance
2.
by anematode 1mo ago
Even worse, in a brownfield codebase that was once fairly light with comments, that's now being subject to these modifications, the insane amounts of commentary around the parts newly touched by AI lead to an excessive emphasis on thos
3.
by anematode 1mo ago
JSON. /s
4.
by anematode 1mo ago
Yes, that's one approach. Another reasonable approach is to get out a mask from the comparison using `vmovmskpd` and use that to look up a shuffle constant, since there are only 16 possibilities. This also works well on NEON, although
5.
by anematode 1mo ago
Great link, nice find!
6.
by anematode 1mo ago
Good question. I personally doubt that the compress instruction is easy to coax compilers into generating, as there are many edge cases to consider. For example, you'll notice here that we perform a full vector store of 8 elements unco
7.
by anematode 1mo ago
Nice post! You can do even a bit better if you're willing to use intrinsics. In particular this kind of operation is well-suited for compress-type operations, available as a first-class operation in at least AVX512, SVE and RVV; you ca
8.
by anematode 1mo ago
That's awesome :) Do you have any links to your guys' work that I could check out?
9.
by anematode 1mo ago
It's a dog-eat-dog world, eh? :) Anyway, I don't see how this relates to the original discussion. Nor do I think the irony is particularly deep: As far as I'm aware, Deep Blue didn't plagiarize Kasparov, and indeed alpha
10.
by anematode 1mo ago
:EmotiTrumpet:
11.
by anematode 1mo ago
> Some times that's appropriate, like if there's a competition for hand-made or human-written something. Indeed, and this is the problem: unlike woodworking, where my power-assisted creation is independent of, or complements, y
12.
by anematode 1mo ago
My implicit request, which I should have made clearer, was for him to delete (or make private) the repository and move on to greater ideals.
13.
by anematode 1mo ago
You'll notice I didn't mention copyright, and that's because I don't think it's relevant. I think your analysis is correct. Keep in mind too that Stockfish – the engine I work on, although I wasn't at the time
14.
by anematode 1mo ago
I'm the OP of the cited thread, ask me anything :P
15.
by anematode 1mo ago
Meanwhile: https://textslashplain.com/2026/08/04/security-is-hard-yall/ > The Cloudflare folks apparently want security issues reported via HackerOne (which wouldn’t let me log in because the Cloudfla
16.
by anematode 2mo ago
Fully agreed. As someone who both loves chess and works on chess engines... these comparisons to chess needs to stop.
17.
by anematode 2mo ago
> deferring int optimizations for later This part made me laugh out loud
18.
by anematode 2mo ago
https://en.wikipedia.org/wiki/Incitement#United_States If done obliquely enough... e.g., just making a tracker and having the ability for people to report broken cameras, you're probably fine in theory. In practic
19.
by anematode 2mo ago
A raw and powerful piece, thanks for sharing. For years I've wanted to get back into self-studying mathematics, not to make serious contributions but just to appreciate its beauty; but the recent observation that I'll never be abl
20.
by anematode 2mo ago
For security?? Why else would they purposely lower cyber refusals in that case.
21.
by anematode 2mo ago
> Do you think there is such a thing as perfect security? No one can "get it right" in the face of arbitrarily high intelligence Why didn't they run the model against the sandbox first? They have effectively unlimited spen
22.
by anematode 2mo ago
Disgusting.
23.
by anematode 2mo ago
Same observation. One thing you can do is close out the session and `claude --resume`
24.
by anematode 2mo ago
It's a "democracy", in which some people (Anthropic and co., and anyone willing to suck up to them) are more equal than others...
25.
by anematode 2mo ago
Demis sold out. "Harnessing AI safely" means nothing when your technology is used to help the US government kill and surveil people.
26.
by anematode 2mo ago
Brilliant! Hadn't seen this technique before.
27.
by anematode 2mo ago
Sure, but you'd use a low-precision approximation for that; you don't need full double precision
28.
by anematode 2mo ago
Agreed, correctly rounded libm functions are great, as long as they don't have miserable worse case behavior (as was famously the case with glibc's pow at one point). One thing I was thinking of doing is manually SLP-vectorizing
29.
by anematode 2mo ago
As someone who has no dog in this race (er, well, I'd like to see both companies crumble!), I think this is a natural response to Anthropic's paternalism, which has always been there but has lately been much more apparent.
30.
by anematode 2mo ago
Exactly!

More ›