Machine learning models do not actually know anything in the traditional sense, but merely predict the next statistically likely word based on vast oceans of stolen internet…

A closer look

Machine learning models do not actually know anything in the traditional sense, but merely predict the next statistically likely word based on vast oceans of stolen internet data, leading to a phenomenon researchers politely call 'hallucinations' where a chatbot confidently fabricates a fake legal precedent or invents a non-existent historical event; the software is not thinking creatively, it is just confidently lying to your face like a sociopathic corporate middle manager, meaning we have essentially spent billions of dollars to invent the ultimate mansplainer, a digital entity that has absolutely no idea what it is talking about but will defend its incorrect output with such flawless grammar that we inevitably believe it anyway