Reading up on Tinker

Thinking Machines Lab fine-tuned a custom model on expert-labeled financial data to outperform frontier LLMs on information-filtering tasks at lower cost.

OpenClaw-RL is a fully asynchronous reinforcement learning framework that trains personalized AI agents from natural conversation feedback and supports real-world agentic RL across terminal, GUI, SWE, and tool-call settings.

Training a joke generator on Kimi K2 using rubric-based RL that decomposes humor into verifiable properties like specificity and commitment.

Co-training a summarizer and generator via RL lets an LLM learn its own text compression scheme, achieving ~9.5% context size with minimal prediction loss.