They said gpt got it wrong on the third question, but what it actually looks like is gpt3 used zero indexing and started from the end of the string instead of the beginning.

While gpt4 began at the start of the string, and also used zero-based indexing.

Makes a lot of sense actually, considering it was trained on a lot of code.

I don't understand why the data are outdated. It didn't know the correct latest version of JDK released either

Is it that hard to keep crawling on a frequent basis?

What is the source? Looks like an interesting article. Some people around me are using it blindly and that is always dangerous. With any tool one needs to know its strengths and weaknesses.

I’d say that it doesn’t matter what kind of indexing you use. The question was about the ordinal and it is 13. So the correct answer is the letter that you get when you start with the first letter, which is ordinal 1, and then count up until you get to 13.
It’s just the rule of our natural language.
The second month is February, no matter how you store it and no matter if it has index 1 or 2.

@Lensflare my 115 IQ answer, vs your 125 IQ answer.

GG Lensflare. I've been schooled.

@Wisecrack thanks. Finally I know my IQ now 🙂

@asgs
It is estimated that gpt-3 was trained in 34 days with an expense of $4.6m.
The problem isn’t about crawling for latest data but the amount of time and resources to train the model with the scraped data.

https://stylefactoryproductions.com/....