Seems like a neat way to check understanding / world-model. Not my idea, saw this post: https://evanthebouncy.medium.com/llm-self-play-on-20-questions-dee7a8c63377
Couldn't this be done in a single instance, by simulating both sides of the conversation?
Seems like a neat way to check understanding / world-model. Not my idea, saw this post: https://evanthebouncy.medium.com/llm-self-play-on-20-questions-dee7a8c63377