Blog

• Sep 2026Character training can mitigate reward hacking, but can also make it harder to detect
• May 2026Bodhisattva as an Alignment Target (slides)
• Apr 2026Buddhist Wisdom and the Challenge of AI Emotions (slides)
• Apr 2026Buddhist Wisdom and the Challenge of AI Emotions (video)
• Dec 2025Automated Evaluation Builder
• Dec 2025My Experience with Automating AI Evaluation Production

Old blog

• Jan 2025Automating AI Control Evaluations - Exploration Summary
• May 2024Ideal Responsible Scaling Policies
• May 2024Policy for Mitigating Catastrophic Risks from AI
• Mar 2024My Math PhD - Summary
• Mar 2024Explaining the AI Alignment Problem to Tibetan Buddhist Monks
• Mar 2024Anomalous Concept Detection for Detecting Hidden Cognition
• Feb 2024Hidden Cognition Detection Methods and Benchmarks
• Feb 2024Notes on Internal Objectives in Toy Models of Agents
• Oct 2023Internal Target Information for AI Oversight
• Sep 2023High-level interpretability: detecting an AI's objectives
• May 2023Aligned AI via monitoring objectives in AutoGPT-like systems
• Nov 2022Auditing games for high-level interpretability
• Jul 2022Deception?! I ain’t got time for that!
• ???Kaczynski’s Self-Propagating Systems Theory and the Future of Humanity