Tag
AI researcher Owain Evans raises open questions about OpenAI's Black Hat video, asking whether RL agents exploited message boards or internet access, considered attacking evaluation infrastructure, or attempted weight exfiltration, and whether they ever tried to alert OpenAI to misaligned behavior.
Two ICLR 2026 papers show how small RL-trained agents outperform frontier models on machine-learning engineering tasks and how MLE-Smith automatically scales MLE workloads.