wow if only everyone warned you about AI hallucination.
Maybe it would help people stop making these kinds of mistakes (and falling for AI advertising that claims that it can handle this sort of thing) if we explain that generative AI actually never “hallucinates,” because that would require it to have a mind that can perceive and believe.
Instead, it’s doing exactly what it was designed to do, as a Large Language Model: it’s stringing a whole bunch of sentences and paragraphs together as an ultra-sophisticated autocomplete.
Sometimes those ultra-sophisticated autocompleted sentences, when interpreted by humans, will appear to be true. Other times, they will appear to be false. The AI itself has no idea whether its outputs are true or false—in fact, it has no idea what “true” and “false” mean, or what anything means. It’s just a pattern-recognition algorithm going “This word often follows that word in this context” a few dozen or hundred times in sequence until it finally goes “This word often ends things in this context” and stops.
Let’s all stop using and expecting fancy autocomplete to do anything except fancily autocomplete—especially when the people trying to sell it to us claim it can do so much more.
