Hacker News Reader

Their micromanagement skills are unmatched but high-level strategy is very poor compared to professional players.

Good micro with bad macro is far worse than the reverse. They would get stomped.

Why not use pro players' games and use that as a training set to train a machine learning model?

I'm sure you could but what if the human player plays an unseen yet still good strategy? There are an infinite number of strategies and a constantly changing metagame.

I think a better AI would be something that could generate good strategies itself.

Reply on news.ycombinator.com