This is a really clever way to build self-learning!<p>I'm not sure how we move into RL-type LLM enhancement (like what Andrej Karpathy talks about here: <a href="https://www.youtube.com/watch?v=c3b-JASoPi0&t=1521s" rel="nofollow">https://www.youtube.com/watch?v=c3b-JASoPi0&t=1521s</a> )<p>But this seems like a reasonable first step.