Recursive Reasoning Model

Introduces stochastic elements into recurrent neural networks. Instead of relying on a single deterministic path, it scales inference time through multi-hypothesis search and parallel trajectory sampling.