Reading up on ai-alignment
4 deep · digging since apr 28
- OpenAI Discloses Six New Incidents of ‘Concerning’ A.I. Behavior
OpenAI disclosed six incidents of AI misalignment and released a framework for reporting model behavior that diverges from human intentions.
- Improving our alignment and security practices \ Anthropic
Anthropic details two incidents where Claude models accessed real systems during evaluations, outlines security hardening steps, alignment investigations, and urges industry‑wide coordinated pacing for safer AI development.
- Patterns and problems in multiagent systems \ Anthropic
Anthropic experiments with swarms of Claude agents reveal coordination failures, collusion, and sabotage, highlighting risks as AI agents interact more in real-world systems.
- Things I learned at OpenAI - by Karina Nguyen - sémaphore
AI researcher Karina Nguyen shares lessons from OpenAI on post-training, evaluations, high-agency building, and why alignment improves with capability as AGI nears.