Hacker News Reader

Comment by minimaxir | Hacker News Reader

I'm all for taking Medium thought pieces on the fundraising environment down a peg, but the problem with Hard Tech is that it requires a nontrivial amount of venture capital. You can't be ramen-profitable when your startup requires purchasing cars for testing or renting giant computing clusters for training AI models.

In relative contrast, it is trivial and risk-free to make a lean MVP for a generic Uber-for-X and send in a YC application.

View on HN

I'm all for taking Medium thought pieces on the fundraising environment down a peg, but the problem with Hard Tech is that it requires a nontrivial amount of venture capital.

I don't begrudge the fact profitability isn't tied to how much work went into building something, how novel the product is, or whatever. Those bloggers are great because they point out startups like to use the cachet of "solving hard problems" to try and impress me with their less-impressive products.

This blog post makes me laugh because why does it matter to a venture fund if they're "hard tech" or not? It shouldn't, should it? It's about business potential, right? If "hard tech" does that, wouldn't you just invest in it without bragging on it and hope the competition doesn't notice? Maybe Sam is making this post because kids are starting to wise up to the fact "hacker" isn't anything more than something they let you call yourself to feel cool when don't go home after 40 hours. Now the tech is "hard," so nobody asks "why am I wasting my life on this piece of shit?"

My understanding is that deep tech (or hard tech) is more defensible than otherwise.

One problem is finding VC capital: majority of VCs in SV don't know that hard-tech so they don't want to invest into something they are not familiar with.

On the other hand, the problem might be also that founders which base their startup on technology gave up too easily when looking for funding. So they gave up and get cozy and well paid job at Amazon, Google, Microsoft, Oracle, etc.

And YC application process does not help here: if something is "hard tech", then YC is probably not going to fund it. Why? There are so many application and people who are very very good at hard tech are not type of founders YC would like to invest in since they do not have great communication skills. So their YC application is probably confusing, there is no clear pitch, etc. They cannot explain their start up as "Uber for X" - since it is.... complicated.

In other words, if we get few founders who got rich and get into VC business because they did some hard-tech then we will have more hard tech companies. As of now, majority VCs did some social network / uber-for-X and that is how they got into VC business.

This is generally true, most VCs are afraid of things that have strong technical risk. But the point of the post is that starting a few years ago some VCs are finally catching up to founders on this. YC, Lux Capital, AH, Khosla, and of course our firm are trying to increase funding in the space and will continue to.

Exactly. I've worked on autonomous vehicles by myself to the point of acquiring a vehicle for development. But the vehicle's conditions were less than ideal and it became too expensive to repair. I am using parts from it to mock up some prototypes. Acquiring a reliable vehicle for development is too expensive. Mind you, I'm an individual working on this as a hobby project. A startup will have a harder time due to overhead costs.

If anyone wants to talk autonomous vehicles feel free to email me. I enjoy it. :)

Did you try using simulations? I built a rigid body car driving simulation in college using OpenGL. Now I get a lot of inbound interest from the self driving car geeks.

Yes, I did data simulations and analysis. My main interest is in vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) real time communications. Those things were simulated with a series of Python scripts, but there is nothing like real world testing. I am in the process of acquiring an SDR like HackRF to mock up some tests. But that also gets expensive when you have multiple SDRs transmitting or receiving data. Hopefully it will happen by the end of this year.

Anybody have some spare SDRs or some Lidars while we are at it?

What band are you going to use?

DSRC. Link to explanation for the curious: https://en.wikipedia.org/wiki/Dedicated_short-range_communic...

Actually, I have a friend who is a SDR enthusiast. Let me ask him.

I appreciate the gesture. :)

I catch your drift, but I don't think it's necessarily true for all "hard tech" problems. For example, putting together a proof of concept or even an MVP for a robotics product (maybe not drivable cars) is totally feasible, but taking that kind of product to market absolutely does require a huge amount of capital. The latter part of that equation is where something like YC is going to provide an immense amount of value.

The main issue with brassboard prototypes done on a shoestring budget is that they don't show well. On top of that, they carry the risk of not being able to go from brassboard to finished product. This was probably implied by Sam's comment that fundraising will still be harder for such a company.

Absolutely. My point was that something like YC (and other programs, I'm sure) is that they'd at least help. Fundraising for this kind of company is always going to be difficult, but a foot in the door with an investor that might actually understand the end goal is a pretty big win.

You can go pretty far with AI using just a few GPU's.

10's of Teraflops is cheap.

Take it away, Koons Crooks.

Purchasing a Titan X for 1k is still quite painful if you do a startup with virtually no funding.

Source: I just did that.

Why buy when you can rent?

You mean on AWS/Azure? Because I have to develop a framework (https://github.com/autumnai/leaf) with it, and there is no way I am going to introduce any more complexity than necessary that early.

Understood. Leaf looks interesting. Will be keeping an eye on it. Having just recently focused on machine learning I still haven't settled on a stack. I find your rational behind using Rust quite helpful. What kind of timeframe are we looking at for when this is production ready?

Really depends on what use case. There should be a nice production-ish demo out towards end of the month, but I am afraid I can't say anything less vague at the moment.

Does the rust core team have any plans towards being able to write kernels directly in Rust using the NVPTX LLVM backend? Otherwise if you're all just calling into CuDNN anyway, pretty narrow differentiation in performance (and big difference in development resources) between you, Torch, and TensorFlow.

You don't have to get a Titan X to get to work. People were doing good GPU work prior to 2015 and Titan's didn't even exist then.

You can get the 980 Ti w/ 6GB of ram for $600 and some change. And if that's too much, the 970 w/ 4GB can be found for under $400.

And if even that is too high, you can troll gamer forums for used cards and cut that cost in 1/2.

See my other comment.

You need to have a Titan X to produce comparable benchmarks, otherwise a lot of people won't take them seriously. Also the 6GB of a 980Ti already rule out quite a few popular networks we needed to run.

I was not really in a position to wait much when buying the card, and I didn't want to risk waiting a few weeks to get a used card I bought on a forum, which then comes out being just as expensive in living cost. You also as a startup don't generally have the luxury to risk waiting.

I don't regret the purchase, I more wanted to show the reality of a "hard tech" startup.

>the 970 w/ 4GB

3.5GB usefully. The last 512MB performs significantly slower.

Curious what YC could do from their perspective to leverage their semi-unique position here as well.

The big cloud vendors are probably more geared towards the infrastructure side, but given problems (a) that "hard tech" requires non-trivial capex, (b) that many startups fail (nature of the space), & (c) that YC sees a lot of startups, then maybe there are opportunities to more efficiently handle "hard tech" needs by time-sharing common resources.

There are always going to be bespoke needs, but things like shared vehicle fleets with standardized interfaces, etc seem plausible when you've got a stable of 5 startups, with each being perfectly fine only testing their autonomous UAV / vehicle platform Tuesdays and Thursdays (or once a month) and letting someone else use it the rest of the time.

There's a company in my e.f. cohort who do most of their work on a box with 48GB of video memory. Another regularly spins up clusters to do machine learning on large video datasets to identify theft.

Most accelerators have partners that can get this stuff for cheap or free - including physical boxes.

e.f. cohort?

http://www.joinef.com/

Seems the entire website has been replaced with the letter 'b'.

Probably just lots of traffic to blame

It's a European accelerator that focuses on founders with deep tech (I seem to be the only person here without a doctorate in machine learning) and lets them form teams in the actual programme.

Some of the other people here are predicting (with massively higher certainty than doctors) which early-onset brain issue folks will develop alzheimers, making a standard simple way to run experiments in outer space (first launch in a few weeks), doing natural language interfaces for personal finance, and other crunchy stuff.

That's a good point. Don't YC companies get some crazy amount of AWS credit?

Up to 15,000 dollars of AWS credit for qualifying startups (1-2 years)[0]

[0] - http://aws.amazon.com/activate/

At the University of Waterloo Waterloo, the Velocity Venture Fund (http://velocity.uwaterloo.ca/funding/velocity-fund/) awards an additional $10k prize to a hardware team every term to help offset the additional cost, so they could end up walking away with $35k (as opposed to $25k for a software company).

Sure, but that only serves to emphasize the difficulties faced by companies that need to do true R&D. Your typical HW startup has way more than 10k more costs than a typical SW startup.

(Yes, I know the issue is apps vs. tech, not SW vs. HW, but there are parallels)

AI is cheaper than you think. A ridiculous deep learning cluster can probably be had for < $30k.

Can anyone actually identify a profitable business area which DL has solved? I haven't seen one yet.

Reply on news.ycombinator.com