''But he has concerns about training generative AI systems on the same outputs they're producing, leading to degraded performance known as "model collapse."
I've had some thoughts on this kind of thing in regard to weather forecasting going away from mathematical predictions to AI - which is well underway atm - I've wondered about using pattern recognition - whilst in a period of rapid change
AI in the sciences seems to work and work very well - eg. Chemistry - but, chemistry doesn't change (as far as I know) --------- but, climate does
will 'model collapse' extend out from LLM's? ----------- I assume that it must
