Hacker News new | ask | show | jobs
by Chestofdraw 3239 days ago
I haven't actually made an 'It's impressive.. but' comment yet but I've wanted to so here's why I'd say that.

Firstly, much like most things a lot of the press releases have been pretty disingenuous. I'm pretty salty about Elon Musks tweet -

> OpenAI first ever to defeat world's best players in competitive eSports. Vastly more complex than traditional board games like chess & Go.

1v1 SF mid is several orders of magnitude less complex than an actual game of dota and probably a lot less complex than chess & Go as well. More than that, 1v1 showcases haven't been a thing for quite a while now for good reasons, the last time they did a 1v1 showcase was at TI4 (in 2014). When they did these the format was best of 3 with the first two games being QoP VS Puck and then the decider would be SF VS SF. This was done because QoP VS Puck is a very mind gamey, strategic, match-up whereas SF VS SF heavily emphasizes mechanical skill where you can snowball off of last hits and denies, giving the match-up more of a sudden death feel. You'd expect a bot to be very good at the SF match-up but struggle with the QoP / Puck match-up where there is a lot more decision space.

To go back to what I said about 1v1 showcases not really being a thing anymore: Back in Dendi's day mid was actually a solo lane but these days in a real game the mid has to constantly be aware of support rotations. The bots super aggressive positioning would be heavily punished even in fairly low MMR pubs so I suspect the pros have deeply ingrained positioning rules that make them hesitate rather than match the bots stance.

All that said, the bot is really impressive and I can't wait for the 5v5 bot next year. It will be super interesting to see what strats it goes for and how much it emphasizes the bots inhuman reaction times / micro skill.