"For now they are going to be hand-selected by me (the area we are starting with requires pretty specific domain expertise). Not sure what we'll do in the future."

Two words: baseband processor.

The world really, really needs a fully open baseband processor.

It really does. But that seems like a mission for a startup, not a research organization. The challenges are more engineering and business than science.

If someone wants to apply with an open baseband processor startup, applications are due Oct 13!

It's true that we need an open baseband processor, but I wonder if this qualifies as the sort of blue-sky research that I'm guessing they'll want to focus on?

My feeling is that the software to accomplish this is already largely there, but the hard part of getting a design to the fab and actually getting anyone to adopt it isn't.

I think there would also be a number of regulatory hurdles to cross, since this would end up being essentially an SDR, and the FCC (for example) already seems to fear wide public availability for those.

They only care if you can make it work on channels it shouldn't, they don't care whether the code that controls it is open or not.

But yes, an open processor could help with a lot of open networking problems, even though it seems more like an engineering matter than research per se.

I thought the issue there was primarily legal, not technical.

We have OsmocomBB, but it's limited to the 2G standard and comes with a ton of usage caveats pertaining to regulatory compliance and legality.

Any helpful teasers on what that domain might be ?

Sam has expressed his desired to get really good AI to become a thing. Since YC does a ton of tech I'd bet $10 it's AI.

But of course this is just my useless speculation :)

25 years is a reasonable number for how long it took deep learning research to begin getting traction. I would think AI would be a strange choice since it seems like there are a lot of research opportunities in academia and industry for this topic, but for the same reason I would say AI research is ripe for disruption.

My bet is on renewable energy/sustainability, agriculture or biotech, with energy being my first choice. Sam has expressed serious concern about powering the future. His first(only?) deviations from the "No Board Seat" policy were two energy companies.

How would funding AI through YC Research be different from what MIRI and FLI do?

Well, they mostly support research designed to prevent harmful AI, which is important but hardly the only or best approach to AI research.

He's also chairman of the board for two nuclear power companies. And YC itself has been branching into biotech and hardware and other science-y companies. Anything seems possible to me.

My money's on nuclear. Advanced nuclear designs regularly claim 25 year timeframes for design, qualification testing, licensing activities, supply chain development, and construction. Transatomic is proposing concepts that require long qualification and licensing developments, many of which could make meaningful progress in a small lab with a $10M initial budget. You have to produce your salt fuel and do materials testing with surrogate fission products (e.g. stable isotopes) over long periods of time. They also have to do hydride transport experiments which can be done on a pretty small scale.

Whoops, I mean UPower, not Transatomic

Exciting times ahead!

A 25-year estimate for how long it will take them to develop an AI that will be able to create biological organisms that utilize nuclear power directly? Sounds reasonable.

Nuclear would also make sense? AI might make more sense becaues nuclear seems like a hard place to do "heretical" research, but what do I know

Nuclear makes sense. The fluid-fueled nuclear company he's the chairman of (Transatomic) is considered at least slightly heretical by the nuclear industry, which is used to its solid fuel water-cooled reactors. They also use used nuclear fuel (i.e. transuranics in nuclear waste) as fuel, which could be called heretical because recycling nuclear waste was once outlawed in the US by Jimmy Carter (but it no longer is illegal).

Trouble is that the amounts of money we're talking about don't go far in nuclear research.

I think $10M can go pretty far in basic feasibility studies needed for the Transatomic reactor. You need a bunch of diffusion couple experiments and hydride stability studies. Lots of furnaces, fume hoods, and surrogate fission products. But you're right that when you get to major component qualification then yeah, $10M is nothing.

I meant UPower, not Transatomic. Whoops!

AI is not really lacking for funding and research activity right now.

Safe AI [1], on the other hand, has [2] running on < $1M and [3] running on $10M (!) from Musk. Throw in Yudkowsky's recently expressed interest in startups [4] and you can guess what I am guessing.

[1] http://blog.samaltman.com/machine-intelligence-part-1

[2] https://intelligence.org/donate/

[3] http://futureoflife.org/AI/AI_beneficial

[4] http://hpmor.com/notes/119/

Now if that "AI" were to be changed to "machine learning" that would indeed make me perk up.

Is there any way that AI does not include machine learning?

I think "we’ll have a process in place to address technology that could be dangerous" is a pretty strong hint, but it would apply about equally well to AI or nuclear, which are already the top guesses in this thread.

@sama - Interesting, I definitely think the toss up between meaningful research and high paying engineering work is very broken. You mentioned how compensation will not be determined based on conferences and papers, do you have any specifics on how you would measure "quality of output"? Will researchers be compensated differently based on said quality?

(As often times the quality of research output is hard to gauge immediately without the greater context of time passing)

Definitely still a work in a progress on how we'll measure it.

ARPA-E (The Advanced Research Projects Agency for Energy, http://arpa-e.energy.gov/) has some interesting models on how to encourage meaningful innovation tied to continued funding. Might be worth a look. Same with DARPA.

Hand-selecting a bunch of famous folks results in IAS Princeton effect which has the criticism of producing nothing significant. Usually the approach of hiring a towering/inspiring figure with an emerging research view works well (like FB did). They bring along a bunch of rookies and experts, and set up research groups quickly and rather smoothly.