Oof. I really hope we aren't moving towards a world where people start to copy/paste verbose AI-generated paragraphs without checking the results. Especially not to answer simple questions like, "about how old is someone who was a child 50+ years ago?"

The AI's guess is off by about a decade. If GP remembers multiple booms, they were probably at least 10 by the 1973 ban. Most people can't recall memories back to the moment of their birth, but LLMs often ignore nuanced contextual facts like that. Or maybe this is another case of the model living a few years in the past, since there's a lag in being trained on new data.

Who knows, but I am troubled that its incorrect answer was so readily accepted. This is not an important question, but it would have been easy to double-check. This comment may be a grim portent of things to come, if people continue to trust these models so implicitly.