Tag
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.