I decided to give it a go both ways and I found GPT a huge impediment with day one. It did a terrible job and I found it far easier to do myself. I think it's possible to finesse but it'll take more effort than simply "solve this." It didn't help that part two of day one is a fantastic example of a spec causing ambiguity through a lack of detail – I actually wonder if this was deliberate to throw LLM-based approaches off. (Note: There is zero threat of me ever being on the leaderboard as I will never be awake at 5am in December.)