Hypothetical proposal to embed undisclosed proofs in LLM training data raises questions about detection and ethics

A thought experiment proposes intentionally inserting mathematical proofs into an LLM's training set so the model would rediscover them, framed as a way to position LLMs as powerful research agents.

The proposal raises the question of how such a technique could be detected without access to the training set, and whether it represents unethical practice.

The author expresses uncertainty about the validity of the idea, questioning whether only bad actors would pursue such an approach.