PointMaze

A new arXiv paper (2609.05461v1) introduces ARC-Bench, a no-leak, fixed-candidate protocol that finds frozen JEPA-style world models rank candidate actions incorrectly, with the top-scored candidate a…

Researchers introduced SegBench-GC, a benchmark testing segmentation invariance in offline goal-conditioned reinforcement learning, finding that artificial trajectory cuts degrade performance: in a Po…

Researchers introduced Rank-Then-Act (RTA), a framework that trains AI using video-derived ordinal signals instead of traditional rewards, achieving state-of-the-art results on benchmarks like PyBoy a…