🏢 AI Alignment Forum

The items that name it 19

• Notes on “We’re in Triage Mode for AI Safety” by Miles Brundage
  articleas classification.key entities2026-10-27
• Monitorability evaluations and policy proposals
  articleas classification.key entities2026-10-22
• Neural nets are exceptionally vulnerable to goal self-hack
  articleas classification.key entities2026-10-21
• The Alignment Problem: Computer Science Meets Analytic Philosophy
  articleas classification.key entities2026-10-18
• Deconfusing Corrigibility
  articleas classification.key entities2026-10-17
• A Simpler and Better Debate Protocol for Scalable Oversight
  articleas classification.key entities2026-10-16
• Preface to the Computational Agent Theory and ASI Alignment Sequence
  articleas classification.key entities2026-10-15
• The Continual Constitutional Convention: How language models revise the documents that govern them
  articleas classification.key entities2026-10-13
• Endogenous Alignment Requires Dependence
  writeupas classification.key entities2026-10-01
• Latent reasoning architectures would likely undermine CoT, our strongest oversight tool
  writeupas classification.key entities2026-09-23
• Why I'm scared of RL
  articleas classification.key entities2026-09-23
• WorkspaceBench: Evaluating Interpretability Methods for the Global Workspace
  writeupas classification.key entities2026-09-23
• [Paper] Stringological sequence prediction III
  paperas classification.key entities2026-09-18
• A Defense of Gradual Disempowerment
  articleas classification.key entities2026-09-17
• Shallow Beliefs: Midtraining does not inoculate against EM from reward hacking
  writeupas organisations2026-09-15
• Op-Ed: I Worked at Google DeepMind. You Should Listen to the Warnings About AI
  articleas classification.key entities2026-09-14
• Latent reasoning architectures would likely undermine CoT, our strongest oversight tool
  writeupas classification.key entities
• Fixed-weight models are adversarially vulnerable: hence misaligned
  articleas classification.key entities
• Continual learning might make your blocking monitors nearly useless
  articleas classification.key entities