Comment by Mc91

LLMs have been getting better - they were all pretty poor for my programming purposes a year or so ago, recently Perplexity (even the non-Pro version) and GPT4 have been helpful, and 4o is even better. I have been posting Leetcode hard problems into 4o and getting sensible outputs, something I didn't even try previously. Sometimes I do have to have it go through a few iterations, and I give it various qualifications (like keep to such-and-such time and space complexity or better). My usual instruction is to make the class or function more and more compact while keeping to the same functionality and time/space complexity.

I got 4o to give me a 33 line, relatively simple and understandable bidirectional BFS Kotlin function for this Leetcode problem which Perplexity (non-Pro) and GPT4 could solve, but not as well as 4o - https://leetcode.com/problems/word-ladder

Of course, even though these are Leetcode hard level problems, they are well-defined and relatively self-contained. I work at a Fortune 100 company and 99% of the time I can pound out the CRUD I do in my sleep - the difficulties I encounter are distractions, the CI server having some problem, the ticket/story I am working out not being fully specified and the PM is MIA that day, all teams are working on the feature at the same time and I need to find out what feature flags to have set and which test headers have been agreed on, the PM has asked me to work on something but some of what he says does not make sense in context so I have to ask for clarification etc. Then there's the meta-game of knowing what to prioritize, with one important component being what will make my manager happy so I get a good yearly review, and what I need to prioritize may differ from what my PM says to prioritize, or even more complexly, what my manager says to prioritize, but doesn't really mean.