@SimonFraser4 Yes! We already order time like that! Imagine having MM:HH. What a nightmare. Better yet, ALL NUMBERS are ordered like that. We don't write our current year as 6202.
@PixelpulseG@SebAaltonen But wouldn't that be an API over another API? Because if so, then one could write your own API that wraps VK/GL/DX. Extremely expensive but would work as a PoC. Sorry if this a dumb question.
@sccplaris Faltou ele estudar economia e política, fazer terapia e participar de debates decentes. Aí ele vai entender que "esquerda" e "direita" é uma simplificação mais prejudicial que útil há décadas (se não séculos).
@fchollet Is there such a knowledge space where it's formally proven to be 100% verifiable and we have devised ways to programmatically enumerate the entirety of the space? The only one that comes to my mind (which could be because I'm drowsy) is math but even then that sounds really hard.
@dhh What did you do to achieve this? "Just" made everything as much data-oriented stuff as possible? So anyone, with or without AI, can modify it with the least amount of friction possible, such as recompiling and rebooting apps/systems/the OS/etc?
Cara needs your help. We have been scraped a third time. This is now a targeted attack meant to cause artists pain.
Scrapers know we're a volunteer project with no financial means to respond. So they say it's legal, believing themselves untouchable.
We're taking action. 1/
@ptrschmdtnlsn Its solution might show up faster but will never be as good as what they could make, if it comes up with a correct solution to begin with. But I believe your point is a bigger factor in the disconnect and worth thinking about.
@ptrschmdtnlsn the better its initial guess and the faster it converges to a solution. I feel like those who usually solve problems "numerically" get a lot of value with AI. Those who usually solve problems "analytically" don't. +
LLMs with powerful harnesses provide, by design, sound output only when the harness includes verification tools. There's no intentional design of bounded time for sound result on coding tasks. Remove the harness and it delivers not even a sound output.
I don't think comparing LLMs non-determinism with compiler's heuristics is correct. Yeah, both can be non-deterministic (and deterministic as well), but one operates over a bounded problem space constructed to be as sound as possible, while the other operates +