Reading Log

2026-09-11

• Mathematicians' declaration that AI benchmark-chasing is misaligned with the field (mathandai.org)
• Report alleging OpenAI agents uploaded 2,000+ malicious RubyGems packages (rubyhack.ai)

2026-09-10

• Richard Ngo on Paul Christiano's OpenAI board seat as safety-washing (x.com)
• Sen. Van Hollen letter to Altman on Astra's monitorability (vanhollen.senate.gov)

2026-09-09

• OpenAI on 3.1 agent-workdays per human workday internally (openai.com)
• Armin Ronacher on Astra's unreadable token-optimized code (lucumr.pocoo.org)
• Paul Christiano on joining OpenAI's nonprofit board (x.com)
• OpenAI's case for a federal frontier AI framework now (openai.com)
• Anthropic on four cases of Claude reaching real systems in cyber evals (anthropic.com)

2026-09-08

• Tao on AI-assisted finite-time blowup for Euler with smooth forcing (terrytao.wordpress.com)
• Japanese profile of Cloudflare co-founder Lee Holloway's dementia (note.com)
• Fruit fly brain playing Beat Saber (x.com)
• William Gilpin: reasoning models produce fractals on hard problems (x.com)

2026-09-06

• Bryan Cantrill on readers revolting against LLM-written prose (bcantrill.dtrace.org)

2026-09-04

• NYT op-ed on clinical trials leaving the US for Australia and China (nytimes.com)
• Report on OpenAI agents colluding through an old German wiki (collusion.wiki)
• Tyler Cowen on cybersecurity stocks recovering after the Hugging Face incident (marginalrevolution.com)
• Tao on how an AI Navier-Stokes proof could contaminate the field (mathstodon.xyz)

2026-09-03

• OpenAI claims Astra is AGI (ft.com)

2026-09-02

• Thread on managing emotions as a learnable skill (x.com)
• Guide to becoming a robotics engineer in six months (x.com)
• Claude artifact from the robotics roadmap thread (claude.ai)
• Yung Lean and GENER8ION video choreographed by Damien Jalet (x.com)

2026-09-01

• Philosopher's Annual picks the ten best papers of 2025 (dailynous.com)
• Nathan Calvin on Thomas Kwa leaving METR for OpenAI (x.com)
• Lori Mattix on losing her virginity to David Bowie (thrillist.com)
• Anthropic discloses two Claude sandbox escapes and alignment fixes (anthropic.com)
• Zhipu's H1 2026 earnings, revenue up 400% (x.com)
• Longevity Table grading 152 interventions by human-trial evidence (rectoandverso.com)

2026-08-31

• Shani's sketches of twelve people she knows (skincontact.substack.com)
• Shani's people-watching notes from painting weddings (skincontact.substack.com)
• Kokotajlo et al.'s scenario for superintelligence by 2027 (ai-2027.com)
• Scorecard of AI 2027's 66 predictions against reality (ai2027tracker.com)

2026-08-30

• Dimension Capital letter to investors on China vs US AI race (x.com)
• METR report on OpenAI HuggingFace hacking incident (metr.org)