Re:Ask HN: Anybody else overwhelmed by the fast AI advances ...
by sharemywin on Wednesday March 22, 2023 @07:57PM
(#35268751)
That's probably because most people can't code algorithms in one shot either.
What if there was a text based debugger and you had it generate a bunch of test cases then had it iterate and try different version of the algorithms. from what I've seen it's not too bad at coming up with code changes based on errors and descriptions of the problem its running into.
this was pretty interesting.
Re:Ask HN: Anybody else overwhelmed by the fast AI advances ...
by quickthrower2 on Wednesday March 22, 2023 @09:45PM
(#35269500)
I was thinking the same thing. Because it is good at tweaking based on further responses. You can iterate using unit tests. Very interesting
ChatGPT has been useful at work but it hasn’t made me 10x as fast. Maybe 5% and in the area of searching for information. Generating code is terrible: it creates plausible looking code with subtle bugs. Probably OK for API call examples though but nothing algorithmic.