[Video] Investor Talk: Machine Learning in Startup Pitches
Investor Talk: Machine Learning in Startup Pitches
Istvan Jonyer, NexStar Partners
Will your startup have machine learning capabilities? Then you need to watch the video below.
Venture capitalist at NexStar Partners shares his views on including machine learning in a startup pitch and the weight founders should give to it, as well as discussing edge cases, using the example of autonomous driving.
Don’t miss the chance to hear from more experts like Istvan at the many tech startup events at Founders Floor in San Jose.
Our Investor
Istvan Jonyer
Venture Capitalist, NexStar Partners
Istvan Jonyer is a seasoned VC investor at NexStar Partners, which recently closed on a $100 million early-growth fund, and invests in sectors such as AR/VR, Machine Learning, AI, next generation infrastructure, mobile security, IoT and video technologies, among others.
Video Transcript
Istvan Jonyer:
So I look at AR, VR and AI machine learning, right. It’s funny how we say AI and machine learning. Like it’s one thing all the time. So, on the VR side, I have a couple of things I’m excited about. So, you asked about the most recent investments. So, my most recent investment in that space is a company called Grid Raster. It’s a seed stage investment I made a few months ago. And what they do is they stream rendered content from the cloud to mobile devices. It goes back to the mobile thesis, but we believe that this device is the key to virtual reality.
So, this is going to be the, the key to virtual reality, right? No one’s going to pay $3,000 for the headset. And when you have something that can do almost everything in your pocket. Except that these things are not powerful enough, to give you the high fidelity experiences. So a great raster runs the high fidelity games on a high powered computer. And then streams the content to the mobile device. And, of course, that’s not very difficult, but it’s very difficult is that they do that with under 20 millisecond motion to photon latency. Given all that, you would expect it to be way longer than that. So they have a number of patents to get that going on the machine learning side.
We talk about, what’s your machine learning thesis? And we can talk about, what makes a good machine learning company. But I think what’s maybe more relevant to this crowd is that, I’ve developed a pet peeve at these pitch events, when companies come out and say, you know, we’re gonna use machine learning to do this and that. And, at this point I’m asking people, if you have machine learning in your pitch, take it out. Don’t say it. I don’t care because, what I care about is that you solve a difficult, important problem and a big problem.
And how you do that at the end of the day matters less. And, saying machine learning isn’t going to get you the next meeting. Saying, big market, difficult problem — that’s going to get you the next meeting. And, the other thing that you have to realize about machine learning, is that machine learning is not something that you want to use. Machine learning is something that you have to use because nothing else works. If you have a problem that’s well defined and there’s a lot of sense for certainty, you can just hand code a solution that works 100% of the time. Machine learning doesn’t do that. Machine learning doesn’t work 100% of the time. It’s like a human brain, right? It makes mistakes. So why would you want to use a technology that’s not 100% and it’s going to make mistakes and you’re going to have to release a product that’s going to make mistakes. Why would you want to do that? Because there is no alternative, right? For human vision and object recognition and voice recognition, there’s just nothing else that we can do. So be prepared to answer the questions of how accurate are you and how are you handling your edge cases? And I don’t want to go into hopefully a whole lecture on this…
Matt Day [Host]:
Yeah, it gets annoying when everyone starts saying, AR, VR, and all this stuff in that stage. Just throw it in there in the pitch thinking, that’s going to get the next meeting – it just never does. You were talking about autonomous driving, I think Bruce, both of you maybe talked about it. When we are we going to have what’s called, I guess would be, is it level five autonomous driving? How many years from now you think before the majority of cars or 51% would be autonomously driven?
Istvan Jonyer:
So, level five, 50% or 50% of the cars on the road. Okay. So, the issue with level five is the edge cases, and excuses just keep popping up and, you can’t, there’s no way to know when the next edge case is going to pop up and you have to have some kind of a solution for it. And when I talked to the experts, they are saying it’s 15 to 20 years out before that’s possible. So first, it needs to be possible before you can build that into cars. Right? And then you still have to wait for the 50% that you’re asking about. I don’t know if that’s the relevant metric, though, because we do have level five at under 25 miles an hour in Mountain View by Google.
Istvan Jonyer:
And so it’s interesting. I don’t know if you guys realize this, but self-driving under 25 miles an hour is basically a solved problem and highway driving is also basically a solved problem. It’s in between that’s difficult. So, under 25 miles an hour, the car just doesn’t move fast enough to cause anything major, you can stop on a dime. Over 75, 70 miles an hour, you’re on a highway, there’s not a whole lot that happens. You just have to keep in the lane and not hit the guy in front of you. That technology was demonstrated in the eighties and seventies. Right? So, the issue is when you go 35, 40 miles an hour in the city where someone steps out in front of you, then what do you do? So that’s where most of the edge cases are, and also the sensor — rain, snow — the car that drives in Mountain View doesn’t drive in Tahoe. And, and all those things.
COWORKING ACCELERATOR
Interested in taking your startup to the next level?