Reinforcement Learning Training Binds to Feedback Environments: High Agent Benchmark Scores Don't Equal Unfamiliar Task Ability
Meta's Muse Wonder scored impressively on programming benchmarks but produced only a red cylinder in a community-initiated 3D lava lamp test. This Leiphone article uses the incident to explain that post-training reinforcement learning shapes capabilities highly dependent on whether the environment provides effective feedback signals, while today's Agent benchmark scores simultaneously reflect configuration differences in execution frameworks and inference budgets.
ImportanceMaterialEvidenceE3 inspectableWrite-upStandard
Meta released Muse Wonder and Muse Wonder Full on October 12, emphasizing coding agents, tool calling, and long-task execution. Wonder Full scored 57.5 on SWE-Marathon and 79.0 on DeepSWE 1.1. The developer community BridgeMind then arranged a 3D programming test, asking the model to create a lava lamp with a transparent glass shell, dynamic lava, and lighting effects; the model produced only a red cylinder. Meta's Chief AI Officer Alexandr Wang responded that the team had not explicitly trained for 3D demo games and that they perhaps should have added relevant training afterward.
Behind this controversy lies a more general technical mechanism: current large models increasingly rely on post-training reinforcement learning to improve task capability. In the pre-training phase, models learn language, code, spatial relationships, and other general knowledge from large-scale data; after entering post-training, reinforcement learning shapes more stable execution strategies in specific task environments. Coding agents are well suited to this kind of training — give the model a code repository with a bug, ask it to locate the problem, modify files, and run tests; the entire process can yield relatively clear feedback through compilation results and unit tests, and the reinforcement learning system uses this to judge which operations are effective and continuously reinforce them.