Essentially, the question asks whether anyone will develop an open source Alphastar that will beat pro players before the end of 2024. The bot should have appropriate APM limitations, which I am leaving somewhat vague, but I assume that Alphastar-level limits are acceptable.
I commit not to trade on this market.
The most promising project in this direction was DI-star by iAsonu
https://github.com/opendilab/DI-star
which is able to take some games from pro players, but not consistently so:
https://www.youtube.com/watch?v=fvQF-24IpXs
Even that took a considerable amount of compute resources (it used 120 Nvidia GPUs for training) training only one match-up and a limited map pool.
In conclusion, I would say that it is quite unrealistic, but leave some probability weight for a team of researchers coming up with a method to drastically reduce the amount of compute or company with a lot of resources improving on AlphaStar.
@Jacek Will this only resolve YES, if it is a variant of AlphaStar, or also if it is a SC2 AI stronger than AlphaStar but with a different approach?
@kiudee interesting, I didn't know about DI-star! I'm surprised it doesn't show up on SC2 AI arena - bots have unlimited APM there, but if it's actually that good, I think it should beat them anyway? (AFAIK they still lose to normal players)
Answering the question: any AI stronger than Alphastar would also count, even if built using different technology.
@Jacek Even though many would like to see DI-star participating in the SC2 AI arena, it is simply not possible because it was trained on a different map pool, a different version and only one matchup.