There are (and should be) many different ways to assess generalization, but you might start with @LakeBrenden’s updates to my late 90’s work. And look at @dileeplearning’s work on how DQN fails in tiny changes to Atari games.
-
-
Our blogs and papers highlight important aspects of generalization in vision, dynamics, and concept learning. https://www.vicarious.com/blog/
@vicariousai1 reply 1 retweet 9 likes -
Replying to @dileeplearning @GaryMarcus and
Do you consider at Vicarious that your tech & team have achieved generalization?
1 reply 0 retweets 0 likes -
Replying to @peremayol @GaryMarcus and
Of course not. We have good generalization in some limited settings. We think we have some of the right ingredients, but a lot more work remains.
4 replies 0 retweets 6 likes -
Replying to @dileeplearning @peremayol and
@MelMitchell1 compare the above to the screenshot from another company I shared recently ... just sayin’.1 reply 0 retweets 0 likes -
Replying to @GaryMarcus @dileeplearning and
Yup, I'm a big fan of
@dileeplearning and@vicariousai
Just to clarify from my previous posts: I wasn't saying that DeepMind doesn't hype their work; they absolutely do. I was just questioning your criticism that they don't have a successful deep RL *commercial product*.2 replies 0 retweets 2 likes -
Replying to @MelMitchell1 @dileeplearning and
They don’t have a successful commercial product that they have sold outside Alphabet. That’s just a fact AFAIK. But I very much read you as saying they weren’t making especially grand claims, which I think they are.
2 replies 0 retweets 0 likes -
-
Replying to @MelMitchell1 @dileeplearning and
Gary Marcus Retweeted Melanie Mitchell
It was your “fifth thought” https://twitter.com/melmitchell1/status/1161776035060301825?s=21 … that colored how I read your thought as a whole; that seemed vastly too charitable to me. I still think that thought is too charitable, but I now understand your thread in a different way.
Gary Marcus added,
Melanie Mitchell @MelMitchell1Fifth thought: Marcus correctly warns of the dangers of "overpromising". But is this fair, with respect to Deep Mind? Have they actually overpromised with respect to their technology? (I did see a quote from Shane Legg a while ago personally predicting AGI by 2020s) 7/Show this thread1 reply 0 retweets 1 like -
Replying to @GaryMarcus @dileeplearning and
In your Wired article, when you talked about "overpromising", it was in the context of overpromising successful commercial products. This is how I meant it here.
1 reply 0 retweets 0 likes
Replying to @MelMitchell1 @dileeplearning and
I do think they are overpromising the practical utility of DRL, but time will tell.
2:25 PM - 20 Aug 2019
0 replies
0 retweets
1 like
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.