Ask HN: How to do independent research in machine learning?

I worked as a freelancer for a while until I couldn't stand working on someone else's project. Having saved up, I have now been studying machine learning full time for a few months, and can keep doing so until the end of the year.

I would like to do research, with a short term goal of solving the ARC challenge by François Chollet (https://www.kaggle.com/c/abstraction-and-reasoning-challenge). I see most professional researchers being guided by imperfect incentives and believe that leaves a lot of opportunity for people like me. Besides, I don't find anything else nearly as important so I have no other choice than to carve my way there!

I have seen John Carmack posting about old machine learning papers, and that got me thinking - what are the places academia isn't paying attention to where one could draw inspiration for alternative approaches?

I know a big part of independent research is that you have a lot of people doing different things so you can increase the odds of one of them finding something worthwhile. This means disagreement in this thread is strongly encouraged. Thank you all!