This blog post appears to be the followup to a very good tweetstorm: twitter.com/fchollet/status/97656387032…

twitter.com/fchollet/status/97678360821… He goes on to defend his work at Google, arguing that they're similar on the surface, but Facebook is truly dangerous where Google is not. I'm not sure I agree with that. Google is quite a bit more distributed across products and platforms, so Facebook has a simpler loop centered around the newsfeed. That said, Google can track a user's behaviour across nearly every website on the internet. Facebook can run these "reinforcement learning on a global scale" experiments through its newsfeed. Google, it seems to me, can run them across the web as a whole.

twitter.com/fchollet/status/97678446524… >Essentially nothing about the threat described applies to Google. Nor Amazon. Nor Apple. >It could apply to Twitter, in principle, but in practice it almost entirely doesn't. I don't believe this for one second. Google does the exact same "algorithmic curation" with its search results. Different people get different results based on internal profiles that Google has built: en.wikipedia.org/wiki/Filter_bubble. Over time that shift in search result content acts upon people in exactly the same way as the Facebook example he describes.

Nah, fchollet just doesn't want pytorch to minimize keras: twitter.com/jekbradbury/status/97661211…