Inside the Secret Meeting Where Mathematicians Struggled to Outsmart AI
Inside the Secret Meeting Where Mathematicians Struggled to Outsmart AI
www.scientificamerican.com
Inside the Secret Meeting Where Mathematicians Struggled to Outsmart AI
12 comments
12 comments
Didn’t read: paywall, but it seems to me that math would be the one thing we should EXPECT AI to excel at.
Have you ever tried to do math with an AI? And I don't mean summing up three numbers or something similarly trivial.
Actually, math problems will probably show best that the current LLMs are just parrots with a large dictionary.
I came to a conclusion about GPT (which is very good with English on most topics) when I asked it how many prime numbers, when divided by 35, leave a remainder of 6. It quickly and confidently said there were none. It hadn't even tried. The correct answer (there's a proof) is: an infinite number.
Two months later it answered that there were 3. Closer ... but no cigar
Not really. They are powered by language models, after all.
Still, much of math is about manipulating symbols. So it is not mind-blowing that a certain understanding of math is there and can be improved on. But math also involves an understanding of space, or geometry. There I wouldn't expect much.
Anyway. My 2 top takeaways from the article:
LLMs no. They are terrible at math.