Gary Marcus

I think I just figured out why @ylecun is so defensive: Samsung’s Starcraft classical-AI heavy bot is (so far) still better than Facebook’s ML system. Results, discussions, links here, h/t ⁦⁦@jackclarkSF⁩ http://www.cs.mun.ca/~dchurchill/starcraftaicomp/2018/aiide/ …

I'm defensive? I don't need to be. But why are you so offensive? CherryPi (the FAIR StarCraft bot) is largely rule based too, with a bit of ML/RL. It's also open source. The Samsung bot is not "still better" than CherryPi since it didn't even exist until now.

ha, i was kidding, but that’s a funny contrast to you yesterday: “If we are talking about Facebook and Google, it's all deep learning ... hard pressed to find any trace of symbolic AI in any operational code.”

You do know that with your "just kidding" you're insulting the team who open sourced FB's StarCraft work, in front of all of your audience, forcing @ylecun to step in to defend their work as you didn't present it fairly? Stop making research a zero sum fight for your own benefit.

but to clarify, @jackclarkSF’s headline said “Hand-written bots beat deep learning bots in annual StarCraft competition”

Yes, which was the thrust of the emphasis of the two bots. The newsletter also includes a discussion of Facebook's bit which uses hybrid of RL and hand-written stuff as well

go hybrids! :)