OpenAI publishes the prompt behind its ETH refutation claim
OpenAI’s released ETH-refutation prompt tells the model “useful rigorous partial progress is preferable to a fabricated solution,” and to say plainly if it failed. Has a model ever handed you a fabricated answer instead of admitting it was stuck?




















