Mo’ turns, mo’ problems – Ian’s Blog

kept by eddie oct 16

Training agentic LLMs with RL suffers off‑policy drift from harness variations, tool‑call formatting, long rollouts, and training‑inference kernel mismatches, needing extra computation or approximations for correct updates.

🎙️Inside a Chinese AI Lab: How MiniMax Builds Open Models
MiniMax researcher Olive Song reveals how they debug RL training, handle model hacking, and prioritize alignment and engineering discipline in open-weight model development.
also on SGLang, vLLM, #reinforcement-learning
Why Open-Source Models Haven't Killed the Big Dogs
Open-source models remain costlier to serve reliably due to infrastructure, utilization, and operational overhead, making hosted APIs from OpenAI and Anthropic more economical despite free model weights.
also on SGLang, vLLM
GitHub - kanishkg/endless-terminals
Endless Terminals is an autonomous pipeline that procedurally generates terminal-use tasks without human annotation for training terminal agents with reinforcement learning.
also on vLLM, #reinforcement-learning
This website has been temporarily rate limited | www.warman.life
Open-weight models from China are commoditizing AI capability, breaking the monopoly moat of US closed labs, prompting protectionist moves that will harm long-term US competitiveness.
also on vLLM, DeepSeek
Notes from inside China's AI labs - by Nathan Lambert
Chinese AI labs leverage cultural humility, student integration, and practical focus to effectively fast-follow frontier models, contrasting with US labs' political conflicts and star-scientist culture.
also on Alibaba, DeepSeek