I'm using it to solve Advent of Code [1] and there a few others doing the same [2]. The training data biases definitely shines trough sometimes but thats mostly when the problem labels (R)ock, (P)aper, (S)cissor as A, B, C. It really wanted to use R,P,S as identifiers. But after you tell it to use the right identifiers it solves the problem.

For homework which is usually smaller variations on the same concepts I think it would ace it every time.

The [2] post is gathering some momentum on HN right now actually [3]. There are many more.

[1] https://blog.nyman.re/2022/12/03/chatgpt-does-advent.html [2] https://note89.github.io/the-advent-of-code-ai-edition/ [3] https://news.ycombinator.com/item?id=33842904