Plain English Papers Creating artificial doubt significantly improves AI math accuracy LLMs are better at math with a "verified reasoning trajectory" An example of intrinsic self-reflection during rStar-Math deep thinking (from the paper).
