"At Amazon we will be investing an order of magnitude more resources towards this problem. With data and computers to match this. This is significant leverage. Hence the change."
The research questions will be in the interest of Amazon, which may not align with the public good (though, admittedly, the logic of the university doesn't always align with the public good). He'll get more leverage, but I'd be surprised if he works on the same problems.
How about updating the academic funding model before everyone goes to industry?
In academia most professors are doing some or most of their projects for the source of funding, which I think is not too much about public good anyway.
I think it is about time for machine learning researchers to go to industry, there isn't that much data in academia. Good public datasets are rare, like ImageNet, and have been basically overfitted to the core. In industry, it is totally another story.
They are usually private datasets and details of the training are not usually disclosed. There could also be duplicated effort by every company competing in the field.
There is also the side that it would make it harder for startups to compete as the progress is made on training on large datasets using huge clusters of machines. Such datasets and compute power is impossible to acquire for small companies.
The fact that Stanford University creates patents from publicly funded research then turns around and sells them to non manufacturing entities makes me wary of pumping any new public money into research at universities. They're clearly corrupt to the core now.
There's a trade-off here: You have to do things that have immediate value to the company (which, at Amazon, corresponds exactly to things that have value to your customers), but you pretty much have the power of the entire internet to do this (AWS). Not to mention more data than most researchers could ever dream of. It's not just monetary.
Since research at Amazon is new, I have no idea how well they're going to handle giving back to the community long-term. But I'm interning with a verification group there and the gist seems to be that if the group wants to give back and there's at least one person who's influential enough to get what they ask for, the group will give back.
I think it's actually problematic that research and industry concerns don't correspond more often. Yeah, it's not our job to be engineers and ship products, but it seems like there are important questions we ought to be answering as researchers that we consider industry ground, but that engineers don't actually have the background to solve. Some of these are totally arbitrary, chosen by the community, deemed "uninteresting" if you want to get published anywhere decent, even though there's plenty of interesting work left.
This sort of thing goes in cycles though. In the late 1990s when I was a postdoc, it seemed everyone in academia who could program or even just knew HTML was running off to a dot.com. A couple of years later most of them came crawling back on their knees begging for their old jobs.
its not just about the data. academia doesnt pay the other workers well and have high turnover. If your a professor trying to run a research lab, your stellar employees wont stay for very long and you'll have high turnover.
HN: Is the future of engineeirng research private? e.g. OpenAI or Amazon/Facebook/Google/Baidu - who seem to have taken many of the world's top AI researchers.
It seems like some of the competing cloud offerings might be slightly ahead in Machine Learning / Computer Vision type stuff out of the box, but I would be AWS has a plan to do more than catch-up.
Amazon is playing catch-up here in machine learning. Comparing the offerings between google cloud and aws, google cloud has a lead.
Not to mention TensorFlow in cloud, you can use their ready-to-use model in voice, translation and image. Does AWS have image tagging? I haven't seen yet.
https://www.youtube.com/watch?v=Ja2hxBAwG_0
I wonder how big his pay increase is. I actually find all the "reasons that aren't $bigpayrise" crap that people go on with in these personal PR things a bit dishonest. I'd bet it's a very large pay increase. I'd be amazed if he isn't getting paid in excess of 350k Which I think is more than CMU pay.
Yeah sorry to be rude and mention money, I just find it more rude to lie and make pretence it's not important when it so clearly is. Maybe I'm wrong, maybe he's such a bad negotiator it's not a pay increase, maybe CMU pay in bushels of gold? But if it isn't a pay increase, it's possible to say so.
Academics are necessarily optimizing for something other than money. Nobody spends 4-7 years obtaining a PhD and then another 5-7 years pursuing tenure if their primary motivation in life is money.
The reality is: Academia can be a real shitty gig when it comes down to it. A vast majority of a professor's life is spent dealing with politics, administrative minutiae, "service", pedantic peer review, teaching, and writing grants. By the time all of that's done, you have no time left for the "fun" parts: doing the research, communicating results and having impact, and seeing its effects on the world. Not to mention the abysmal funding climate, changing role of universities, and the fact that many fields (esp. in CS) are now being pushed further, faster, and better by industry...
There's a mass exodus going on right now. At least 25% of the professors I know (I'm friends with many) have either left or are thinking about leaving academia.
CS is lucky in that it's relatively cheap to study. Other than compute time, all most people need is a good coffee maker and a white board. Biology professors need to bring in an order of magnitude more grant money to keep their labs running. As a result, most of them do very little of their own research; they may look at data or plan an experiment with a grad student or postdoc, but the vast majority of their time is spent raising money or reviewing other people's work.
Yes I regret every day leaving science as I love it. In an ideal world I would still be a professional scientist, but we don't live in an ideal world so you have to make the best of what opportunities you have - I certainly can't complain about my life.
My only advice is seize the chances that come your way and make the most of them. It is all too easy to stay in the warm cocoon of academia even when you know you should leave.
I got out a little over a year ago. I was a BME PhD and did a lot of software engineering and signal analysis of brain signals for brain-computer interfaces. I was a neurology professor for a few years (after a 4 year neurosurgery postdoc), and decided academia wasn't for me. I now work at a trading firm as a quant researcher/trading system software engineer. I have never been happier, but definitely miss the science part of research (i.e., NOT grant writing, politics, etc). I have been doing part time consulting work on Upwork focusing primarily on helping with biomedical signal analysis and software/app design. I still get to be involved in that niche, but now I have money left to feed my kids after all my student loans have been paid. We are actually able to buy a house, which was going to be more than a decade away had I stayed in academia (and got NIH funding, and tenure, etc). I guess I value financial security and that of my kids too much to sacrifice my life for doing science.
As someone who recently left what was ostensibly a dream job (a tenure-track position at a top university), I largely agree .
Being able to start to avoid the "service" of donating my time to support the paywalled journal system and actually enjoying open source development again (even though I get to do less of it) is the best part.
You're off the mark here. Most people who do a PhD and go on to becomes a professor don't have money as an important goal, because doing a PhD and then getting tenure takes ~12-15 years, and nobody can afford that kind of opportunity cost. If money were one of their most important considerations they would have left for industry before their PhD.
Nonetheless, top professors get paid decently (although obviously they could make 5x in industry). Michael Jordan at UC Berkeley makes ~320k. Newly tenured CS profs make ~160k at UC Berkeley. Source: https://ucannualwage.ucop.edu/wage/
It's true that people doing PhD's aren't monetarily driven but you can easily imagine that incentives change once they start a family and have to worry about things like a mortgage, college tuition for their kids, etc. If you can get a significant pay raise while maintaining some semblance of research autonomy AND have greater impact, why not go for it?
As I mentioned, a CS professor with enough seniority to have a family to raise is going to be pulling a 200k+/year salary, more than enough to comfortably raise a family, even in the Bay Area (probably not SF).
Something I didn't mention earlier is also the fact that if they really needed extra money they can easily consult on the side, as well, which is totally okay in academia (unlike most employees at companies).
> they can easily consult on the side, as well, which is totally okay in academia (unlike most employees at companies).
It is very rare to be outright prohibited from consulting on the side as an employee. I currently do exactly that (work for a company that sells consultants and developers to its clients on contract) and we are only prohibited from doing the same on the side if it is a conflict of interest. I've never heard of someone going to their supervisor with a lead and being told they could not pursue it.
Yes, it is rare to be outright prohibited. But it will definitely be looked down upon by managers. It is also culturally uncommon. Very few Google engineers are consulting on the side, for example.
It is extremely common for CS professors to consult. And there are absolutely no conflict of interest restrictions. CS professors will consult in their direct area of research.
What I'm saying is I know people who've started families while they were postdocs or even phd students. Academics can start families at any level of seniority.
And if the country's policy is "well of course you're struggling to support your family, it's your own fault for not waiting until you achieve tenure at average age 39" [1] well I can understand academics leaving academia for industry. The average age of first childbirth in the US is 26, with steep rises in complications beyond age 35 [2].
Unless you're a man planning to have a child with a woman 10 years your junior, getting tenure before having a child doesn't sound like a great idea to me.
The thing is, it can be both money AND a better opportunity to make an impact that led to him making that decision.
However, it probably feels wrong to write a blogpost and say "I did it for the money". So people conveniently leave the money part out but I think we can all agree that it's probably definitely a non-trivial factor in their decision. It just happens to make more sense to tactfully focus a blog post on the better opportunity at hand rather than the money.
Yeah it probably is more than $350k, but who's going to put that on a blog. Oh look how much I am making now. People are supposed to read between the lines and I think the "it was an offer that I could not turn down" gives a pretty good hint in the right direction.
I would wager moving from being a tenured private university professor in Pittsburgh to Seattle to work for Amazon for money, even in a VP level role, would be a poor choice.
Unfortunately, moving because you believe Amazon will throw resources at a problem shows a distinct lack of familiarity with the "frugality" aspect of Amazon's culture.
He will probably be a principle or senior principle and will have a lot of freedom over what his group does, as long as he can justify that it helps the company in some way (which really shouldn't be any more difficult than justifying funding to ${government-organization}).
Freedom to control what your group works on versus resources (high quality staff, compute, etc.) to effectively pursue that work are very different things.
Fair enough. My experience didn't take me to AWS, but in retail it was a problem for entire tier 1 services to get resources with VP support, let alone new development/research initiatives.
Perhaps Amazon has changed, or AWS is simply different/better in this regard.
The compensation was ok at the time, but starting to lag their peers by the time I left (within a year of leaving my comp had a little more than doubled). My understanding is that they've made an effort to catch up there. Regardless, Pittsburgh is a staggeringly cheap place to live (I grew up there) while Seattle is decidedly not (I live there now). Unless he's tripled his comp, he'll be making effectively less. That wasn't my point, though, I assume he's not moving for the money.
I was actually talking about resources to dedicate to research.
What makes you think Amazon will be willing to supply compute (assuming they actually have substantial excess cycles to sell, particularly on GPUs, FPGAs, or ASICs) to internal development efforts? Simply having the resource doesn't mean making it available. Actually making appropriate resources available to people trying to solve problems would run completely counter to my experience there.
I don't have any inside information or anything like that, but...they are hiring him to extend his previous research on large-scale machine learning systems.
As the lab head, I imagine he won't have a hard time getting resources--they must have given him something like a startup package to lure him there.
As a big name, they have a strong incentive to keep him happy: if he were to leave in a justifiable huff (e.g., "they promised me tons of compute and all I got were these Fire Phones"), it would make recruiting much harder.
Even when just looking at the title, it's safe to assume that CMU is some kind of place, more specifically, an organization and/or institution. Reading the first sentence of the article -- "I'm leaving CMU to join Amazon" -- brings even more clarity, that "Leaving CMU" refers to a changing of occupation. The next sentence talks about machine learning, so it's safe to say that the author is in the AI/machine learning space, and that CMU is a place that, like Amazon, has some expertise in it.
At that point, if you haven't already guessed that CMU means "Carnegie Mellon University" -- an acronym that's not quite "MIT" but still well-known in engineering circles -- you'll get the definition in the very first Google search result, even for just "CMU"
It seems to assume people who read a CS professor's blog or a site called Hacker News will know Carnegie Mellon University, given that it has consistently hosted a top-5 CS program for the last 20+ years.
US News ranks it #23 of domestic schools. It has top-10 graduate programs in Compute, Mechanical, Electrical, and Environmental engineering with an overall engineering school ranking of #5. It has a top-10 fine arts program (#6). It is weaker (not in the top 25) for pure sciences and applied math, although it's #11 for discrete math and combinatorics likely due to the CS program. It's business school ranks #7 for undergrad and has top-10 graduate programs for information systems, operations, and logistics.
So, by and large, it's actually a fantastic university if you want to be an engineer, a computer scientist, an artist, or a business person. It certainly isn't "VEEERY poor" in all other fields.
I wonder what words you'd use to describe an average university, seeing as you use 'VEEERY poor' for one that's apparently rank 80ish (out of the thousands if not tens of thousands that exist).
They work well together. The drama students like wearing costumes (for example, robot costumes) and even the dumbest drama student is still way better than the typical robot AI. It's all an act.
The research questions will be in the interest of Amazon, which may not align with the public good (though, admittedly, the logic of the university doesn't always align with the public good). He'll get more leverage, but I'd be surprised if he works on the same problems.
How about updating the academic funding model before everyone goes to industry?