LogFAQs > #978755115

darkmaian23 posted...

I'm surprised so many people find ChatGPT useful. So-called "hallucinations" are rampant. Like, I see people saying they are having ChatGPT teach them new things. If you don't already know about the thing, you won't spot the errors and will assimilate false information. Summarizing long documents is also problematic for the same reason. If you haven't read the document, how many of the mistakes will you catch? Helping you rewrite a paragraph or two makes sense, or maybe help with form letters, but having it write papers for you is a terrible idea from a learning and personal growth standpoint.

EPR-radar posted...

This. As far as I can tell, the AI chatbots are the most efficient way known to provide high volumes of plausible bullshit.

It will be remarkable to see what fraction of the economy is driven by efficient production of high volumes of plausible bullshit.

But I can't see chatbots being useful for anything that requires understanding.

Hallucinations do happen, but I wouldnt necessarily call them rampant. At least, not in the more modern GPTs. Also, understanding why hallucinations even happen can help you use AI in a way to minimize them.

For me, hallucinations happen when I am more than ~5 or 6 questions deep on a specific subject. They also happen when I am not specific with my instructions.
You can minimize these problems by creating a detailed, organized and very specific prompt. The more limitations you give it, the better. Also making sure to repeat the original context with further questions helps keep it on track.

Its extremely helpful and accurate in my use cases