Simon Willison

simonwillison.net

03 August 2025 22:47

Here are my notes on XBai o4, the latest 32.8B open weights LLM to come out of an AI lab in China, this time from new-to-me MetaStone AI simonwillison.net/2025/Aug/3/x...

Patrick Dubroy

dubroy.com

16 July 2025 08:00

TIL: Exponentially-segmented arrays → github.com/pdubroy/til...

Sung Kim

sungkim.bsky.social

12 May 2025 08:23

• Detailed Technical Report: primeintellect.ai/intellect-2 • INTELLECT-2 on Hugging Face: huggingface.co/collections/... • Chat Interface to try it out: chat.primeintellect.ai • Blog: www.primeintellect.ai/blog/intelle...

Sung Kim

sungkim.bsky.social

12 May 2025 08:18

Prime Intellect releases INTELLECT-2 (open-sourced first 32B parameter model trained via globally distributed reinforcement learning), which demonstrates that globally decentralized RL works, per them. They also open sourced: - PRIME-RL - SHARDCAST - TOPLOC Validators

Richard McElreath 🐈‍⬛

rmcelreath.bsky.social

12 May 2025 08:18

Said another way: If interactions important but not identified due to confounding, them average treatment effect in your sample may be valid, but you can't project that effect to population of interest, because sample distribution of interactions likely different than population dist