Episode Details

Today's papers: - Before They Can Solve: Predicting Post-Training Coding-Agent Performance from Base Models: https://arxiv.org/abs/2610.10478v1 - Open-MMUnlearning: Unifying Methods and Evaluation for MLLM Unlearning: https://arxiv.org/abs/2610.10358v1 - Long-WAM: Scaling the Context of World-Action Models: https://arxiv.org/abs/2610.10528v1 - TaoD2C-Bench: Benchmarking MLLMs for Industrial UI Code Generation Beyond Visual Fidelity: https://arxiv.org/abs/2610.10374v1 - RaReCache: Bridging the Gap in Cross-Model KV Cache Reuse via Rank disagreement-based Selective Recomputation: https://arxiv.org/abs/2610.11358v1 This podcast is from Colin Davis (colin-davis.com) using Claude & Elevenlabs.