From: Bob Berwick To: "Jeffrey E." <[email protected]> Subject: Re: Date: Sat, 05 Sep 2015 14:03:29 +0000 Hi Jeffrey, Ah very cool. So let me try a few out. I guess it partly (! of course!) depends on how much we get to stick into the box. I'm fine with prize money, mostly because as you know from Marvin, there's no way to get funding for "real" problems like these or those that Marvin studies, nsf & darpa and toyota (as you just saw) only care about self-driving cars. So I'd be overjoyed to get some way to fund new students.... Challenge I. Discriminating language from noise. This is doable. Partha Niyogi nearly did it. A six-month effort. Challenge 2. Ah, depends on the sentences and what `coherent' means. Does Noam mean `screwed up in some way'? We're not so far from that. Further than 1. That system I showed you that can isolate diffi factors in *syntax* as a small set of constraints can do some of that. Recall Shimon Ullman's `rigidity principle' about the interpretation of points moving in the visual field - same coherence idea. Challenges 3 and 4 are much harder, we'd have to know more about 'meaning' and `inference'. I wouldn't be happy with statistical curve fitting here. Finally - though I will drop this, because your time is too valuable and the problems you and Noam pose are the important ones, to me — what was your estimate of how much that twitter-in-love entrepeneur would have lost if he did honestly follow his twitter-predictor during the first week of this month? I guess I am still baffled about the "market" for their product unless it's as you say, the market for the new, shiny toys - which is just about salesmanship, not science. Anyway, I'll drop it - I was just curious as why/how *any* of that social media stuff would work. I suppose they will say that they have 300 M users, not *all* of them will be lying... and one can sort the liars from the non-liars....but so what? best wishes bob On Sep 5, 2015, at 8:20