More links:

1. A critical look at Lean4 https://www.lesswrong.com/posts/zHhGa3iQBQcdSifFL/a-critical-look-at-lean4-1

2. Harness Learning Enables Generalizable Test-Time Adaptation https://arxiv.org/abs/2609.35738

3. Evo2: Evolving the Evolver through Weight-Harness Co-Evolution https://arxiv.org/abs/2610.13025

4. From AlphaGo to AlphaZero, reinforcement learning produced strategies that surprised expert players. AlphaZero learned from experience at scale, without human demonstrations. AlphaNIST brings this philosophy to contact-rich, high-precision robotic assembly in simulation. An end-to-end policy learns how to complete the task, without prescribed manipulation stages. https://alpha-nist.github.io/

5. Can LLMs produce original research worth reading? MathArena gave four agents 48 hours and a $500 research budget each. GPT-6 Astra and GPT-6.1 Sol delivered technically strong but off-topic posts; GLM-5.3 found a modest insight but analyzed it poorly. The authors judged only Opus-5.5’s post useful, though it still needed editing. It proposed a new benchmark—and performed worst on it. https://matharena.ai/llm_blog_posts/

6. FrontierML: a challenging, end-to-end machine learning benchmark https://www.frontierml.org/blog/

7. 100+ reactions to 100+ solutions https://proofsandprompts.com/2026/10/08/100-reactions-to-100-solutions/

8. Linus Torvalds Working On A Hefty Linux Kernel Improvement With Claude Code https://www.phoronix.com/news/Linus-Torvalds-Claude-Code

9. How biological synapses self-assemble gradient learning https://www.pnas.org/doi/10.1073/pnas.2606245123