Poster wants it to play Jeopardy, not process text.

Care to enlighten me?

I don't think he does. Larger models are definitely better at not hallucinating. Enough that they are good at answering questions on popular topics.

Smaller models, not so much.

Not sure if you're correct, as the market is betting trillions of dollars on these LLMs, hoping that they'll be close to what the OP had expected to happen in this case.