anematode

6 ms·

61.
by anematode 4mo ago
Edit: Another developer seems to have found a legitimate speedup with Fable in an optimization loop. It's a nice idea, actually, and I'm duly impressed.
62.
by anematode 4mo ago
No, because Wikimedia isn't responsible for the behavior of its editors.
63.
by anematode 4mo ago
Yup, I suspect that's what's going on
64.
by anematode 4mo ago
It's utterly bonkers. Hopefully the model weights get leaked. Then we can claim it's public domain or, at the very least, distill it and then release it for free.
65.
by anematode 4mo ago
Not impressed so far, to be honest. I'm having it try to optimize Stockfish in a loop (on xhigh mode) with a benchmarking oracle; even after giving it specific hints ("consider whether we're prefetching Y optimally, can we ma
66.
by anematode 4mo ago
> there are thousands of tests and CI won't let things to merge if tests are failing. That's not what a clean setup means... I mean good separation of concerns, established invariants, etc.
67.
by anematode 4mo ago
Does this not indicate that the project might not be structured in an appropriate way that allows incrementally adding features?
68.
by anematode 4mo ago
The next model will fix this.
69.
by anematode 4mo ago
Noooo, the real thieves are the Chinese AI companies which used Anthropic/OpenAI model output as training data. American AI companies can do no wrong. /s
70.
by anematode 4mo ago
A difficult problem domain, yes... so where was the staged rollout? The penetration testing? Why did it take so long to disable the bot after reports of the exploit surfaced? etc.
71.
by anematode 4mo ago
Hm? JavaScript BigInts are arbitrary precision, and you need to use methods like BigInt.asIntN(64, a) to convert them to 64 bits
72.
by anematode 4mo ago
> The compartmentalisation that must be required by the scientists and engineers to reconcile with the fact that their work being used to bomb and kill people must be crazy. I think about a related question pretty often: What proportion
73.
by anematode 5mo ago
> This is what fifteen years of standards work on a five-letter keyword ['volatile'] looks like. How many letters are in the word 'volatile' ?
74.
by anematode 5mo ago
To the contrary, Nova Lake, coming out this year, will have it.
75.
by anematode 5mo ago
Ah yes, ClickUp, the issue tracker that was so sloppy there wasn't a single person at my company who was upset about our switching to Linear, and which used several gigabytes of RAM on the main page. Excited to see 100x of this.
76.
by anematode 5mo ago
Notably, Bun is not a JS engine. JavaScriptCore is the JS engine. Bun is just a complicated wrapper around it.
77.
by anematode 5mo ago
Relax, and enjoy drawing on some eggs.
78.
by anematode 5mo ago
Well, scheduled by whom? :)
79.
by anematode 5mo ago
Quite the mentality...
80.
by anematode 5mo ago
> You're subsidizing everyone else if you're not trying to get the best loyalty program. Quite the mentality...
81.
by anematode 5mo ago
My trick for when the desktop app is buggy is I have Claude decompile it and fix the issue. I have a series of a few patches (I think this is one of them)
82.
AI Wellbeing: Measuring and improving the functional pleasure and pain of AIs(ai-wellbeing.org)
2 points by anematode 5mo ago|0 comments
83.
by anematode 5mo ago
I stand by it. I'm not including all Google employees, ofc – there are some fantastic projects coming out of there – just the people working on their AI systems which will be accessible to the government with (effectively) no oversight
84.
by anematode 5mo ago
Who could have seen this one coming. From yesterday: https://www.cbsnews.com/news/google-ai-pentagon-classified-u... ("Hundreds of Google workers urge CEO to refuse classified AI work with Pentagon"). Any AI
85.
by anematode 5mo ago
Maybe... but I would love if 1% of the investment in AI were redirected to the mathematics education and professional research that would allow progress on any of these problems...
86.
by anematode 6mo ago
Neither does the Collatz conjecture, Fermat's last theorem, .... (Of course, those problems are on another plane than this one.)
87.
by anematode 6mo ago
The mobilizers, in this case, are almost entirely the people leading and deploying AI rather than any anti-AI agitators. When someone hears from these leaders that there will be a white-collar "bloodbath", then sees enshittifica
88.
by anematode 6mo ago
I don't think this is technically possible without something like homomorphic encryption, which poses too large of a runtime cost for usage in LLMs
89.
by anematode 6mo ago
Am I alone in thinking this stuff is nuts? (Currently half way through the article, btw.) Analyzing "emotion" in the model is completely anthropocentric. If we indulge in the idea that LLMs of sufficient complexity can be consciou
90.
by anematode 6mo ago
My favorite (admittedly not super useful) trick in this domain is that sbb eax, eax breaks the dependency on the previous value of eax (just like xor and sub ) and only depends on the carry flag. arm64 is less obtuse and just gives y

More ›