paper / brianchristian / May 12 / failedReward Model Interpretability via Optimal and Pessimal Tokens
paper / brianchristian / May 12 / failedSelf-Blinding and Counterfactual Self-Simulation Mitigate Biases and Sycophancy in Large Language Models
paper / brianchristian / May 12 / failedAI Assistance Reduces Persistence and Hurts Independent Performance