matrix07012 @matrix@gameliberty.club

@alyx

@tomie he was doing it before moving there, guess he's spiritually Romanian

@arc DeepSeek R1 is probably the best you can get locally (you can get even the full model, but it's like 700GB). That thinking is a feature, it helps with accuracy quite a bit.
I think that unfortunately you simply can't fit a good enough model into such small amount of parameters.

@nyanide It's supposed to be basically a self aware mockery of various gaming subcultures, but it's mostly communists posting genuine takes.

ok, scratch that, Grok is just acting like the last two messages just don't exist

@8757d9c788ddfa02b91056961aa1bced110fa7bd1716af2540c7d013aad337e5 Obviously yeah, I mean great as in comparison to other models

Imo we have hit or are close to hitting the LLM plateau or at least period of stagnation.
OpenAI fumbled massively.
Sonnet 3.7 seems like a small improvement and maybe even a regression.
DeepSeek R1 is great, Grok 3 is great, however thinking models in general seem to be a band aid.

Monster Hunter Wilds PC - Profound Perf Problems Must Be Addressed
https://www.youtube.com/watch?v=0yhacyXcizA