[cadamsdotcom]

Agreed on the IP point. Strongly agreed on not wanting to go to court in the current US political climate.

On the rest, especially the confidently incorrect argument.. Not. so. much.

Firstly, models are stochastic parrots, but that truth is irrelevant because they're useful stochastic parrots.

Second, hallucinations and confidently incorrect outputs may yet be a thing of the past, so we should keep an open mind. It's possible that mechanistic interpretability research (a fancy term for "understanding what the model is thinking as it produces your response") will allow the UI to warn the user that uncertainty was detected in the model's response.

Unfortunately none of that matters because the IP point is a blocker. Bummer.