But the LLM is almost literally "the collective wisdom of real-world experts" rather than a single one. Obviously it's not at that level yet, but I don't see a clear upper bound to it getting there. I'm not very familiar with athletes, but from what I know, regardless of the "human premium", if the AI coach helps their competitors perform better, they'll need to adapt or "perish".
Coaching, especially at elite levels, isn't about information. At that point the human connection to a coach with championship experience is better than pretty much any conceivable chatbot.
I'm confused, why would you insist on using an LLM for decision making? It's not
"the collective wisdom", it's a probabilistic language model. I imagine LLMs are better suited as a last step. I'd want a proper model to infer a decision, and an LLM to transform it to a human-friendly-formatted advice.
