Learning ML, there was a high emphasis on the error part of things as most of the course was on minimizing errors. After ChatGPT, there is a weird anthropomorphization going on, where it's all about hallucinations, alignment and what not.
replyWe have something that is statistical in nature so there should never been any expectation of error-free results/actions. The value has always been about discerning trends or the cost of errors being way lower than any good result.
Statistical learned indexes can exist in explainable tech such as a database.
replyIn 2017 Google was writing papers about it. Then something changed.
I don't think it was the tech. It was a realization around the power and societal impact.