OK: following, my thoughts on Google Brain and similar institutions. Disclaimer: this is my opinion, as someone who doesn't need their funding, and has built a career without needing to flatter them. (An anti-disclaimer, if you will.)https://twitter.com/TehRaio/status/1017612699692359681 …
-
-
"Deep learning implements the renormalization group!" Yeah, I heard that too. If you have a system where information is organized spatially, is it really a surprise that the neurons group information together spatially?
এই থ্রেডটি দেখান -
I'd get invited to meetings at Google Research, or wherever. They had security like crazy—worse than a hedge fund. A security guard would follow you to the bathroom.
এই থ্রেডটি দেখান -
Each scientist at my "grade"—i.e., the equivalent of a junior faculty member, someone who should be out on the edge of knowledge—was, instead, managing a team of ten people doing graduate student descent.
এই থ্রেডটি দেখান -
Google can beat University of Kansas for the sole reason that they can hire ten times more graduate students per researcher. The difference, of course, is that a graduate student at UK has the chance to do something intellectually significant. Not true at GR.
এই থ্রেডটি দেখান -
They had no idea what they were doing. They had the manpower (word chosen advisedly) to apply deep learning to anything, simulating the Schrodinger equation, drug design, anything. Their main goal was to find the scientific field they could have the maximum impact on.
এই থ্রেডটি দেখান -
I've visited probably fifty Universities. I love it. Everywhere I go, I get new ideas. It's one of the best features of my job. There's one exception: commercial "research" labs.
এই থ্রেডটি দেখান -
If you want to build machines that monitor people and sell them more ads faster, go for it. If you want to find problem where you can take a working-class job, model the man or woman who does it, and build a net to put them out of a job without compensation, be my guest.
এই থ্রেডটি দেখান -
Have we done science with something Google Research has built? Absolutely. We have a great paper coming out where we use word2vec to help build a theory of puzzle solving.
এই থ্রেডটি দেখান -
But we could have built a system of equal utility ourselves. There's zero intellectual contribution there. I'm not joking, and I'll go head-to-head with anyone who says I am.
এই থ্রেডটি দেখান -
I got a nice cold-call from a top-flight Masters' student in CS, as I do sometimes (please keep them coming, I can pay). I flew him out and we started working on a problem in the emergence of social cooperation. He wanted to do DL.
এই থ্রেডটি দেখান -
Two weeks in we were a step beyond what Google Brain was doing. I don't mean technically—they had amazing YouTube videos of sprites in a landscape. I *do* mean intellectually. Their demos were like 2018 meets something out of the 1980s.
এই থ্রেডটি দেখান -
They said they did social science, but it was nothing of the sort. It was homo economicus spread out over 50 GPUs. At best, a devastating proof-by-example of need for academia. Buy a copy of Bowles and Ginits, Cooperative Species, and you'll learn more than they did, in a week.
এই থ্রেডটি দেখান -
Can you do cool research at Google Brain? Honest answer: no. You will be on the cutting edge of machine learning, yes—an engineering discipline whose basic goals are set by large corporations. But you will not be a scientist.
এই থ্রেডটি দেখান -
I get that you may need to make money. You can make a lot there, and all the jobs at Renaissance Technology are taken. Go for it—you have all my respect. Academia sucks.
এই থ্রেডটি দেখান -
But if you want, at some point in your flourishing career, with your mind and your soul, to join the two-thousand year old parade of intellectual progress, you are not going to do it at Google. Certainly not at Facebook.
এই থ্রেডটি দেখান -
If you want to do that, I have a suggestion. It's not the only path, by any means, and I've had amazing fellow-travellers who haven't. But here it is.
এই থ্রেডটি দেখান -
Go to graduate school. Do a PhD. With us, here at CMU/SDS, if you like—but we're not the only place that does computational social or cognitive science. You won't get paid much, but you will mentors who legitimately care about the development of your mind.
এই থ্রেডটি দেখান -
It's difficult to overestimate the difference between a good PhD program and industry. It is literally shameful, if you're a good PhD advisor, to interfere with the intellectual development of a student. At Google, it's a business plan.
এই থ্রেডটি দেখান -
None of this is a joke. This is ten years of experience. Graduate school applications are coming up in the Fall. Think about it. Make sure you're getting a good deal (you shouldn't go into debt for a PhD, and you should get healthcare).
এই থ্রেডটি দেখান -
In short: corporate "research" is a business proposition. Whatever true intellectual progress comes out of there happens in spite of management. Given how good these companies are at monitoring their employees, that gap is now miniscule.
এই থ্রেডটি দেখান -
Last anecdote, then I'm done. We visited Google Research, arranged by a contact. The people were unbelievably smart. We brainstormed all sorts of wonderful things to work on. The last day of the meeting, the academics were like, OK! Let's go to the pub! Let's hash this out!
এই থ্রেডটি দেখান -
Their response: this was vacation for us. We're behind on our real work. We have to work this weekend. (Not "we feel guilty", but "we have to".) For the academics in the room, this *was* work. Suddenly, I realized that this was vacation for them.
এই থ্রেডটি দেখান
কথা-বার্তা শেষ
নতুন কথা-বার্তা -
লোড হতে বেশ কিছুক্ষণ সময় নিচ্ছে।
টুইটার তার ক্ষমতার বাইরে চলে গেছে বা কোনো সাময়িক সমস্যার সম্মুখীন হয়েছে আবার চেষ্টা করুন বা আরও তথ্যের জন্য টুইটারের স্থিতি দেখুন।