This is the kind of multi-tasking I want to see in general world models. Applying domain expertise in rl to strengthen overall architecture. Gonna be interesting to see if this approach scales to other domains. https://www.reddit.com/user/MasterScrat