Weirdly I feel lot more accepting of LLMs in this type of environment than in making actual products. Point is doing things fast and correct enough. So in someways LLM is just one more tool.
With products I want actual correctness. And not something thrown away.
Given what I understand about the nature of competitive programming competitions, using an LLM seems kind of like using a calculator in an arithmetic competition (if such a thing existed) or a dictionary in a spelling bee.
I feel like it’s more like using an electronic dictionary in a spelling bee that already allowed you to use a paper dictionary. All it really does is demonstrate that the format isn’t suited to be a competition in the first place.
Which is why I think it’s great they dropped the competitive part and have just made it an advent calendar. Much better that way.
These contests are about memorizing common patterns and banging out code quickly. Outsourcing that to an LLM defeats the point. You can say it's a stupid contest format, and that's fine.
(I did a couple of these in college, though we didn't practice outside of competition so we weren't especially good at it.)
We’re starting to get to a point where the ai can generate better code than your average developer, though. Maybe not a great developer yet, but a lot of products are written by average developers.