top | item 45061414

The limits of ChatGPT: a week of interaction and AI research

4 points| MTarver | 7 months ago |marktarver.com | reply

3 comments

order
[+] MTarver|7 months ago|reply
What happens when a scientist spends a week in continuous, serious dialogue with an artificial intelligence — and finds their health, their thinking, and even their loyalties changed?
[+] attogram|7 months ago|reply
Wow that's a deep dive.

And wow 2: "The conversation was so disturbing, that I archived it and went on holiday"

[+] MTarver|7 months ago|reply
It wasn't at all what I was expecting. The thread actually opened with a conversation on post-WWII British aircraft manufacturing of all things and got in automated reasoning. ChatGPT made a clever remark on group theory which prompted the paper. From then on the rest followed very quickly. The whole thing was mind-blowing really and archiving it and going away was the only thing to do. I was getting sleep problems.

As regards the research I think the best thing to do in these cases is to feed ChatGPT the paper and start where we left off. This seems to work but it is a bit too early to gauge how well this will work.