Some links:
1. Reframing LessWrong-style decision theory as "commitment theory" https://eliasschmied.substack.com/p/reframing-lesswrong-style-decision
2. "A personal update on infinite ethics" https://eliasschmied.substack.com/p/a-personal-update-on-infinite-ethics
3. How can the invisible be made visible? In his lecture as part of the new Ars Mathematica series, Fields Medal winner Peter Scholze takes the audience on a journey into the world of condensed mathematics. With clear language and palpable enthusiasm, he shows how abstract thinking opens up new spaces of understanding and why maths is sometimes actually a bit of magic. https://www.youtube.com/watch?v=uHJizsbVmhk
4. A closed-loop platform pairing a ranking-based protein language model with robotic experimentation improved cytochrome P450 BM3 activity 57-fold in five weeks and five design–build–test cycles. https://www.nature.com/articles/s41467-026-76264-2
5. Frontis-MA1 / OpenMLE-Evo: A system for automating machine-learning research by post-training a coding model specifically to draft, improve, debug, and recombine ML programs, then deploying it within a long-running evolutionary search process. https://arxiv.org/abs/2607.28568
6. Scaling Automated Post-Training: Intology’s report on Locus, an automated research system that uses large-scale parallel experimentation and specialized infrastructure to post-train language models, outperforming general coding-agent baselines on PostTrainBench and its higher-compute extension. https://intology.ai/blog/scaling-automated-post-training
7. Dispatch from Anthropic v. Department of War Summary Judgment Motion Hearing https://www.lesswrong.com/posts/jGEXLKyGtXyiYa7ac/dispatch-from-anthropic-v-department-of-war-summary-judgment
8. Serious cyber vulnerability disclosures kept climbing in July https://epoch.ai/data-insights/cve-severity-spike-july-2026
9. OpenAI: Apple is getting this wrong https://openai.com/index/apple-is-getting-this-wrong/