Chatgpt is not good at these kinds of problems. Heck, any solution and/or answer it gives should be checked manually if you intend to apply it somewhere. It may seem smart most of the time, but you’ve just hit one of its bigger weaknesses.
The worst part is the way it states these thing as fact instead of just “so you should be…”. Definitely tricks people. Obviously this one is easy to spot, but it does the same thing no matter whether it is right or wrong, and if you have no idea about a topic you’ll just believe what you’re reading is correct.
I only ever use this for quick things that I have a pretty strong grasp on but still need a little push over the hill to solve a particular problem. Almost always I get a response that’s only half correct at best, but something inside the response I can pull out and use to solve the issue I’m having.
I think that’s a pretty good approach. I think it would benefit all to see ChatGPT more like an approximation/guessing machine. It often hits the mark, or even gets really close. Its bigger hallucinations are frequently hilarious.
you need to override that overly helpful system message they put in chatgpt.
“be concise, do not explain, only provide the answer”
Sometimes the explanation is the easiest way to catch it making logical errors.
Reducing context when there is an error is a good idea too.
Upvote for using ISO8601. Cheers
It actually had the right logic/approach to solving the problem with steps, but undermined it’s own solution when it solved the first step already accounting for months and days. If a human was to provide this level of detail, the first step should have come out to 8 because 2004 - 1996 is 8. Then subtracting 1 would have correctly yielded 7 as the answer.
Math checks out
