Reinforcement Learning for Reasoning in LLMs with One Training Example

Turns out that https://www.fortressofdoors.com/four-magic-words/ was right and all you need to do in training is have the LLM meditate on a single example.