[projektir]

> Often times at the bleeding edge of research, that one postdoc is literally the only person in the world who can do both tasks.

I thought we just said earlier how easily replaceable this post-doc is.

Not buying it. Bleeding edge of research doesn't really exist, it just doesn't move fast enough to have any sort of "bleeding edge". It's slowed down by lack of money and poor management and too much bureaucracy far more than it is by someone not working long hours. As has always been true with these things.

This is why startups will, sadly, end up beating academic science over time. Because startups are bleeding edge.

"It has to be this way!" is a very difficult claim to verify, you can't expect others merely to accept it, you must prove it, and I have seen very little evidence so far, including from my own experience with people who did research at university.

[johncolanduoni]

> Not buying it. Bleeding edge of research doesn't really exist, it just doesn't move fast enough to have any sort of "bleeding edge". It's slowed down by lack of money and poor management and too much bureaucracy far more than it is by someone not working long hours. As has always been true with these things.

You're thinking of bleeding edge as "new drug that at least shows up in some pop-sci stuff". People working in labs come up with new methods of doing X in situation Y all the time because X and Y can both be crazy specific to a certain line of inquiry. A thousand of those lines of inquiry will likely be explored before anyone outside of their extremely specific area of study notices. Your sampling bias is pretty irrelevant to the process.

As a concrete example, maybe you need to apply a novel algorithm to a high frequency data source (e.g. a single photon counting module). So you need to program an FPGA with a deserializer to do the processing (in lieu of wasting money on a DAQ that can pump data into your computer at a few GHz sampling rate), and only one postdoc understands the algorithm and FPGAs well enough to do it. Does that sound like a crazy, unusual, made-up scenario?

> This is why startups will, sadly, end up beating academic science over time. Because startups are bleeding edge.

Have you seen how hard it is for startups that actually work on bleeding-edge scientific work to take off? Is that not something that Y-Combinator (for example) has specifically been looking for a more hybridized model for? Most research done at university is a very risky bet even for startups, especially if you want not just a result but a marketable result.

Want some concrete examples of the system working? Take a look at the envelope that quantum information research has pushed in the last 5 years. On the theoretical computer science side, you've got people like Scott Aronson answering significant open questions at a dizzying rate (read his blog for some basic, well-explained evidence). Meanwhile you've got groups like Martinis'[1] making the first quantum computer that can accurately simulate a different quantum system. All with a huge amount of collaboration in a social network of scientists that spans the globe.

Where's your evidence that startups are the hammer-screwdriver-impact-driver combo that solves all of humanity's intellectual problems with but one institution? Because that's a much stronger claim than "academia does moonshot research that wouldn't get done otherwise and its not the scientific equivalent of another day another CRUD app".

[1]: http://web.physics.ucsb.edu/~martinisgroup/publications.shtm...

[projektir]

Bleeding edge means time is important. Because time being important is the only thing that would make the argument. Since time is not important, it's not bleeding edge. So there's no actual rush, and any claimed one is artificial.

There's no justification for rushing in science in general, in fact, and all the rushing for publications and other rubbish is going to make all the research worse, not better. It should be a careful, deliberate process. If you are working 16 hours and are constantly worried about what your professor thinks of you, I am forced to be wary of your science.

> only one postdoc understands the algorithm and FPGAs well enough to do it.

I'm not going to go into problems of poor knowledge transfer obviously present here (bus factor 1?), but if this was the case, the post-doc would have negotiation power and this entire situation wouldn't be an issue. Clearly the post-doc is extremely replaceable, as the letter in the article implies, so you should have like 10 of those. If not, why are you threatening the post-doc? This sounds like a situation of not enough people, not too much people.

This asymmetry concerns me, and I am rather confused by it.

> Have you seen how hard it is for startups that actually work on bleeding-edge scientific work to take off?

It's hard, and overall, I would say it's worse (hence the sadly). But it's a lot better than the nonsense that academia is engaging in right now. It at least provides some competition and puts pressure on academia. If academia doesn't fix its act, eventually, private will have to take over. It's the same situation as it'd be nice if all the big, well-equipped companies would create and promote the electric car, but if they won't, someone else will have to do it, even if it's less preferable.

Something tells me that the big name post-docs can actually dictate their own terms and are not the same post-docs we're talking about. I'm having a hard time imaging that post-doc getting emails like this.

There's lots of ML research on the startup side right now, I wouldn't be surprised if the demand for quantum research will rise with time. The main advantage academia has isn't the system, but the government money... which is not at all indicative of the system working well. ahem government contractor companies ahem And seems like private can also grab government money.

We got a lot of good results from the "guy just goes on his own and is left alone by everyone and just sits there and studies alchemy for a bit" system, too. Consider how much was accomplished before with very little effort and how much people are working now, and how many people. I understand the problems are harder now, you need better equipment, etc., but that doesn't mean I'm not going to call a spade a spade. I've seen the time of so many post-docs wasted that I'm not going to just ignore that.

You can't expect everyone to just accept that, it needs to be justified. That's what I mean by "not buying it". It may be true, but the evidence is not there, not there at all.

[johncolanduoni]

> Bleeding edge means time is important. Because time being important is the only thing that would make the argument. Since time is not important, it's not bleeding edge.

That seems like a very artificial definition of bleeding edge. I've only ever heard it as a description of working in an area where a lot of important aspects are unexplored and using any praxis based on what we've already gleaned is risky. I'd like to know where you're sourcing that "time is important" condition.

> I'm not going to go into problems of poor knowledge transfer obviously present here (bus factor 1?),

The head of the lab may very well not know how to use FPGAs (nothing about what I described entails the lab was researching any kind of computer engineering), and the postdoc might have even taught himself how to do it just for this task.

> but if this was the case, the post-doc would have negotiation power and this entire situation wouldn't be an issue

It is still an issue, because you can find another smart postdoc replacement that will teach himself how to do it without raising a stink about his hours spent working for the next 10 times he needs to do something crucial. Will it take more time for this one subgoal? Yes, but replacing someone who is constantly starting salary negotiations every time he thinks he has some leverage is in the end likely to be a time-saver.

> It's hard, and overall, I would say it's worse (hence the sadly). But it's a lot better than the nonsense that academia is engaging in right now. It at least provides some competition and puts pressure on academia.

So it's worse, but it's better? You've failed to explain (at all) how you know startups can solve these problems when they have no track record of solving anything like them. Without that explanation, your justifications smack of motivated reasoning and isolated demands of rigor.

> It at least provides some competition and puts pressure on academia. If academia doesn't fix its act, eventually, private will have to take over.

Academia already has plenty of competition internally; it's one way that the limited number of tenure and postdoc positions has been quite beneficial. You've also again failed to even mention how private will take over, and how it'll be any better once it does.

> There's lots of ML research on the startup side right now, I wouldn't be surprised if the demand for quantum research will rise with time. The main advantage academia has isn't the system, but the government money... which is not at all indicative of the system working well.

Yes, once ML research started producing useful results in reasonably predictable timeframes. I agree quantum research will also enter the startup world once it starts producing useful results in reasonably predictable timeframes, and that change of gears will accelerate development. But how do you think the field is going to develop to the point where it can be depended on for revenue?

> You can't expect everyone to just accept that, it needs to be justified. That's what I mean by "not buying it". It may be true, but the evidence is not there, not there at all.

I provided a concrete example of an area of research in academia moving forward at breakneck speed in an area that doesn't yet have much value to businesses (their involvement has all been within their own moonshot, academically modeled research divisions). But you've provided zero examples of startups excelling in such an environment, and I'm just supposed to buy that?