←back to thread

165 points fzliu | 1 comments | | HN request time: 0.203s | source
Show context
lynguist ◴[] No.41853965[source]
Chatgpt can write a solution for this problem if we go up to 10^10. 10^16 is too hard!

I don’t understand this problem (I didn’t tackle it myself) but wanted to see how quickly chatgpt could solve it and how far along it would go.

First it made a naive solution that would work until 10^6. Then we used that output to verify improved versions.

And we managed to improve it until 10^10 only. (Staying within a minute timeout.)

It did a DP approach (it suggested itself). I suggested to try numba and numpy. And with that it managed until 10^10. I think it’s still brute forcing it and one might leverage much better techniques in order to reach 10^16.

replies(1): >>41854943 #
1. kevinventullo ◴[] No.41854943[source]
It hadn’t occurred to me that PE is a pretty good test for ChatGPT or SOTA LLM’s in general. Even if explicit solutions for earlier problems do appear somewhere on the web, it’s probably a fairly safe bet that the latest version of ChatGPT did not include the latest Project Euler problem in its training data.