Fine tuning an open source model to detect LLM attacks. Started out with this game where you try to get the LLM to give up the secret key: https://www.integrated.io/