right now people either trust AI way too much to NEVER make mistakes, or don't trust it at all after it makes ONE mistake. in September CNN put out a story about an AI chatbot providing our military with hallucinated info during the Iran war. an intelligence analyst had been using AI to help write up a report on a ship, but somewhere in that process the chatbot got the ship's cargo wrong, and the report it produced falsely claimed that a Chinese vessel in the Middle East was carrying parts for a nuclear weapons program. but the analyst trusted what the AI gave them and submitted the report normally, and it moved through the military chain the way a report of this magnitude normally would. nobody questioned where it came from until right before they were planning to board the ship. that's when somebody took a harder look at the report, realized an AI had written it, and shut the operation down. to me, that's over-trusting AI at its absolute worst. but a story like this usually pushes people too far the other way, to where they stop trusting AI at all. if you look at what actually went wrong, though, the chatbot made one mistake, and the bigger failure was that nobody checked its work until it was almost too late. that's why i think people need to learn how to afford AI the grace they afford themselves. when you get something wrong at work, you don't decide you're useless, and you never stop checking your own work. the model is going to be right on most things and wrong on some, the same as a person. the difference is that it produces so much, so fast, that its mistakes add up quicker when nobody's reading. getting the best results out of AI is just as much a human process problem as it is a technical one.