I've been tracking a distinction between well-specified versus non well-specified tasks for almost as long as I've been thinking seriously about AI, fwiw (since 2016 or so). I've been tracking it the entire time LLMs have existed in a pretty serious way. E.g. this is essentially what my PhD was about (starting in 2019) which I was interested in pursuing in part because I was noticing this delta between artificial and biological intelligence. I don't think "well-specified tasks" is quite the right name of the category (I don't think we know what the true name is), and I've used different words in the past, as well as noticed different aspects and refined more of what I mean over the years, but the core distinction has been live for me for a long time.