Nothing to Go On

This week: Self-play, context language models, santa clause, employee disclosure with AI, influence and binding.

No Single Safeguard

This week: Reliability theory for AI Control, auditing sabotage bench, autobenchmark, optimal timing for superintelligence, serious AI product.

Pain, Apparently

This week: Debunkable beliefs, pain responses in llm’s, AI changing labour demand, physical capabilities, MoE as efficient retrievers

Something Spooked the Experts

This week: Reliable weak-to-strong monitoring, corrupt plans, stellar colosseum, cognitive virus, new polling, Whoopi.

Hidden Curriculum

This week: Mechanism design for alignment and control, you are what you read, cheating and whistleblowing in autonomous research swarms, confidence shortcut, course design in the age of AI.

The Lathe

This week: If God Arrived As a Chatbot, Late-Stage Pressure States in Long-Horizon Tool-Use Agents, AI Control safety case, Questionnaire, The Coming Storm

In the Wild

This week: METR’s brief independent investigation of agents’ behavior, The Hugging Face incident, Breaking ArrowCloak, Prime Agent, maturation process

Boundaries

This week: open-weight genome language model safeguards, watermark localization, AI guardrail survival, stealing reasoning traces, A US strategy to prevent the creation of mirror life

Large Worlds

This Week: Genome language models, weak verifiers, agentic inequality, cognitive commons, cooperation in large worlds, cooperative AI, genesys open models