“Prime Intellect plans to train models to use the recursive language model (RLM) harness over the next few months.”