alt Hacker NewsI think your analogy is good but I don't believe modern LLMs use Lean or any lean-like structure in their proofs. At least recent open source ones like DeepSeek can do advanced math without it (maybe the most cutting edge ones are doing it I can't say).
They are most likely using them in training. I doubt their IMO team are show ponies