Can GRPO be used for multi-turn RL?
Meta panicked by Deepseek
DMs with 20+ years of experience. What aspects of the game do you still struggle with?
Billions in proprietary AI? No more.
Is there any RPG that is set in the "call of cthulhu" universe, but not Horror-centered?
All the AI models seem to be utilitarian-leaning - Why aren't they Kantian?
How do mereological nihilists respond to gunk?
Is science possible without some degree of empiricism?
Are science and empiricism linked? Can you believe in one but not the other?
[D] Titans: a new seminal architectural development?
If physicalism and moral realism hold, what are morals made of?
Why do most vector databases use a NoSQL format rather than SQL?
Former OpenAI employee Miles Brundage: "o1 is just an LLM though, no reasoning infrastructure. The reasoning is in the chain of thought." Current OpenAI employee roon: "Miles literally knows what o1 does."
Do I lose out on anything from doing pacificst for my very first playthrough?
OSR ruleset: Spells and Blades (one page, two sides + index card character sheet)
I created one page (two sides) OSR game as an experiment and challenge for myself. Let me know what you think! Did I capture essence of OSR?
Would Oedipus-style prophecy or time travel imply fatalism or be problematic for compatibilism?
Cyberpunk settings or settings with megacorps on a galactic scale
Top Agent only 27% away from degree-holding humans on GAIA (General AI Assistant) benchmark (created with Yann LeCun)
Introducing SmallThinker-3B-Preview. An o1-like reasoning SLM!
DeepSeek-R1-Lite-Preview seems to beat DeepSeek V3 on multiple benchmarks, so why is V3 getting so much more hype?
Is there an inverse of the ARC AGI challenge? Something very hard for humans and very easy for LLMs
Why neural networs work ?
DeepSeek V3 was made with synthetic data for coding and math. They used distillation from R1(reasoner model). Also they implemented novel Multi-Token Prediction technique
How are attention heads able to attend to features stored non-linearly?
[D] How are attention heads able to attend to features stored non-linearly?
