A Tiny, Exact Lab for Judge-Policy Self-Play
A fully enumerable toy experiment on evaluator drift, policy collapse, and the ceiling on self-improvement in judge-policy co-evolution.
Why RLHF Can't Scale: Understanding the Fundamental Limitations
Examining why RLHF faces fundamental limitations across scalability, human judgment quality, reward models, and governance that constrain the development of more capable AI …
What Works in Synthetic Alignment: Evidence and Scorecard
The verdict is in. We deliver a scorecard on synthetic alignment, assessing which of RLHF's limitations have been solved and which remain, backed by six key empirical insights.
The Path Forward: Five Critical Research Frontiers
Exploring five critical research frontiers: meta-alignment, post-deployment adaptation, breaking the iteration ceiling, judge bias auditing, and co-evolution dynamics.