An Omitted Mode Is a Rare Rule: The Sampling-Verification Danger Law in Continuous Code World Models
Crucial for ML researchers working on verification and safety of AI-generated code models in continuous control systems.
AI Summary
Researchers identify a sampling-verification danger law in continuous code world models where accepted models can miss critical events despite reproducing sampled transitions.
Excerpt
In the Code World Model paradigm an LLM synthesizes an executable world model that a classical planner searches, and the model is accepted when it reproduces sampled transitions. We ask what that acceptance certifies in continuous control. We define the pipeline's danger as an expected risk and isolate its exact factor: the probability that N i.i.d. gate rollouts all miss a critical event of probability r is exactly (1-r)^N; an independent acceptance sample adds its budget to the exponent. On th