Comment Useful, but verify important stuff (Score 1) 247
I use ChatGPT quite a lot. I like to ask off-the-wall questions just to see how it responds. It's also fun to have it write a short story as a demo for someone else who's unfamiliar. ("Write a 200 word story about the time [name] saved the day using only [two common unrelated objects]") But I've also asked very complex questions about network configurations, troubleshooting obscure hardware failures, and horticulture. I generally get accurate and workable answers straight off. Sometimes I just describe a problem I'm having, without asking any specific question, and it almost always is able to infer what my question would be and give a correct answer.
I do find the bootlicking cheerfulness a little annoying. It reminds me of Ford Prefect's short-circuited security bot.
I have had a couple of instances where it was clearly hallucinating, and it doubled down when I challenged it. Here's an example: I saw a "Dedicated to:" slide at the end of a TV episode so I asked ChatGPT what it was about. In response I got a biography of a fictional woman, and a paragraph describing her importance to the show. When I responded that I was pretty sure that was not accurate, the second response I got was an apology, followed by a statement that there was no dedication. When I said there definitely was, I got a biography of a completely different fictional woman. At this point, I rewound the video, took a screen grab of the dedication (which was to a man) and uploaded that. ChatGPT apologized again, and finally gave me a couple of paragraphs about the actual person.
I do some programming, but have not used ChatGPT in a programming context so I can't speak to that.