Episode Details

Today's papers: - From Interpretability to Control: Insights from Six Years of the TrustNLP Workshop - Reinforcing Step-level Reasoning for Effective Self-Correction in LLMs - Rule of Thumb: Explaining Artificial Intelligence Systems using Partial Information - VERDICT: Training-Free Step-Wise Verification of Multimodal Reasoning via Disagreement-Aware Consensus - Rubric Dropout: A Simple Way to Mitigate Reward Hacking in Rubric-as-Reward RL This podcast is from Colin Davis (colin-davis.com) using Claude & Elevenlabs.