[WhitneyLand]

I’ll take that as it happens so infrequently with GPT4 you have no illustrative prompts that can be shared.

There have not been tons of papers written about this.

You seem to be conflating papers about GPT4 as a solver with it as a math tutor. It’s a completely different problem space.

[godelski]

Or you can check the front page:

https://news.ycombinator.com/item?id=38845878

(older but similar) https://news.ycombinator.com/item?id=37904047

[WhitneyLand]

I don’t get the relevance those seem to be security related?

My main point consistently has been that GPT4 can be an invaluable resource specifically for learning math subjects.

I am not aware of any papers, studying people using it as a conversational tutor for learning math and having problems with hallucinations.

[godelski]

> I don’t get the relevance those seem to be security related?

And?

> My main point consistently has been that GPT4 can be an invaluable resource specifically for learning math subjects.

This can also be true. I use it a lot. Don't confuse openly discussing limitations with calling it a pile of shit. No need to have only two extremes.

> I am not aware of any papers, studying people using it as a conversational tutor for learning math and having problems with hallucinations.

Very bad faith requirement. Unless you have good evidence that GPT hallucinates in many domains (as exemplified by said security report) and NOT math tutoring. If you have this really strong evidence that math tutoring is specifically unique then I suggest writing a paper. I'll help if you really can do it and be happy to give you first author and be proven wrong. But a much easier explanation is that math tutoring is not unique to GPT with regards of generating hallucinations. If you truly believe you do need a extremely specific example, you may need to pull the wool off your eyes. But I'm hoping you don't and are just arguing.