3 ms·

So reinforcement learning…

• diggan 4y ago
  Not really. Although there is a state (the source code) and a action being performed (suggest improvement then suggest edits) there is no reward so it doesn't really optimize itself. But in some test runs it added code to evaluate how long time it took to run the full flow and started adding heuristics for choosing best edits, so I guess you could say it started doing reinforcement learning in some cases.