Reading a UCB PhD thesis from 1974 republished as an MIT Technical Report. The general rule of thumb now is: Humans emit clue, when clueful. LLMs regurgitate and distort said clue. They do NOT emit NEW clues. For me to ask them to contribute to original research would be largely pointless... the computational shotgun machine performs poorly when actual understanding is needed, or they mean-reverted to something in the training set because they couldn't process the query contextually. I will never shake the fact that none of Claude Opus, Haiku or Sonnet, or Qwen for that matter, could figure out correct answers from basic questions I asked of them about the code in a git checkout, which was PUBLICLY AVAILABLE, right in front of me. Training MY ASS! The cheap(*) boxes of stochastic tricks did not impress me at that moment, and they still don't. Prepare for financial mayhem. (* Special Discount due to Negative Externalities! Yes! "Disruptive innovation" my ASS!)