The biggest jump in technology has already happened. They already trained on all available data. The models are roughly as good as they're going to get. We will never truly be able to trust an AI model not to hallucinate stuff when applied to critical applications. The current tech will always require a human in the loop. Running more models to help other models is a hack and not a true solution to the problem of models being wrong.

Yeah, this post was not my finest work.