I am not an AI engineer, but my intuition tells me if we could ever clean up the @#$& datasets these LLMs are trained on and give them coherent, non-contradictory training, we would be shocked by what they could do.

I suspect 90% of the criticism of AIs is because people are underestimating them.