[jerf]

"How many hours ago was june 1st 1972": https://chat.openai.com/share/3e3bb478-f7b1-49e1-81f9-235afe...

Payload of the answer: "So, approximately 455,796 hours have passed since June 1, 1972, as of February 16, 2024."

Today (as I write this) minus 455,796 hours: https://www.timeanddate.com/date/dateadded.html?m1=2&d1=16&y...

Which is Thursday, February 17, 1972. Since it just did the year and ignored the month you were asking about. (I accidentally deleted my first conversation instead of sharing it but it gave this answer twice.)

The real question though isn't whether ChatGPT is wrong, the real question is, can you detect it? That's going to be the important question here going forward.

[simonw]

Try "How many hours ago was June 1st 1972? Use Python"

https://chat.openai.com/share/d5264b01-3be5-4d59-b175-34ad90...

"June 1st, 1972 was approximately 453,304 hours ago."

Code it generated and ran:

    from datetime import datetime

    # Current date and time
    now = datetime.now()

    # Date and time for June 1st, 1972
    then = datetime(1972, 6, 1)

    # Calculate the difference in hours

    difference = (now - then).total_seconds() / 3600

This illustrates my biggest complaint about ChatGPT right now: the amount of knowledge and experience you need to have to use it really effectively is extremely deep.

How are regular users meant to know that they should hint it to use Code Interpreter mode for a question like this?

[swozey]

You're 100% right, I use it daily like I said and I really have NO idea what all the other options are. I tested a plugin or two and it caused problems with something so I disabled it.

I need to dig more into how much more I can do.

[swozey]

Ha, I didn't actually test it and I was curious if the answer would deviate at all from reality. Thanks for doing the math. Thankfully I know what bad rust looks like, but I'll totally get suckered for the wrong math.