[Pinkthinker]

I think that maybe we are too harsh in expecting LLMs to be perfect. If they are based off of human input that is incorrect then we might propagate such errors. But they will still be quicker and much more reliable than most people. Isn’t this good enough? After all, we are willing to accept flaws in people, even including the president. I suspect that the way forward will be to progressively clean the LLM input data as each error gets identified.