And yet the best models still give bad answers a high percentage of the time, lie, and make up links, and possess nothing remotely resembling human intelligence. All of the popular models work by recognizing patterns, so it is not surprising that it might be helpful with a problem which can be solved by recognizing patterns. Although ironically the model itself likely has no actual understanding of what it solved.
“Although ironically the model itself likely has no actual understanding of what it solved.“
Yet.
Terminator was a documentary.
That’s not too surprising. The models are built to to find patterns. So on-task for this problem. But humans hallucinate connections all the time - ever heard of Alex Jones? That’s what you’re mostly seeing in what you’re reporting regarding poor AI performance. Hallucinations. Humans are still better at recognizing when they’ve made an unrealistic connection (well most). - but we make the connection in the first place, then correct ourselves.
Right now (and for all time, I would presume), *you* are the gut-check on AI result to call BS. I know that when I see BS in an AI result I tell it to the AI, and I often get a much better answer back.
It’s a product. If a car company is putting out cars that catch on fire you don’t close down all car manufactures. You go after folks building ‘the pinto’...