Hacker News new | comments | show | ask | jobs | submit login
Ex-Googlers left secretive AI unit to form Groq with Palihapitiya (cnbc.com)
76 points by coloneltcb on Apr 21, 2017 | hide | past | web | favorite | 16 comments



Amazing not only in the speedup, but also in saving electricity.

Deep learning is changing the field of AI and I expect that with a few more breakthroughs in training algorithms and hardware everything will start to change, hopefully for the better. I wonder when we will see ultra low power neural chips in cellphones.

In the 1980s, my team at SAIC built special purpose neural hardware: Harvard memory architecture designed to run backprop quickly. We got 5 megaflops out of our boards and we were happy enough.


I don't think this will have anything to do with training, since it's the TPU team and trying to beat nVidia at floating-point matrix multiplication doesn't seem like a great idea. Run-time can generally be done with integer operations, which is near dirt-cheap these days - I'd think the only interesting things would be operating at server scales or so low-power that you're getting 30FPS on your classification/segmentation what-have-you on your smartphone without impacting battery life. My guess is the former, though I'm not sure what the market size is for companies that need to evaluate their machine learning systems at speeds fast enough to merit specialized hardware.


I don't quite understand how these xooglers can take any of that knowledge with them and start a new company. Wouldn't everything they do be patent encumbered? Sounds like a big fat unknown liability to me. Not to mention non competes. What am I missing?


I recently read a Vox article that explained how California courts don't enforce contracts that limit employee mobility (and partially explains why behavior like this is possible):

http://www.vox.com/new-money/2017/2/13/14580874/google-self-...


Patent law =/= contract law. They're not referring to non-competes


I don't know about non-competes, but you can't patent math. You can patent systems that use math, but not the math itself. If the application is different enough they might be fine?

I'd be interested to hear from somebody who knows more.


> I don't know about non-competes, but you can't patent math.

If you call it software you can.


And in Europe, if you call it a "system" (software+hardware) you can (which those chips are).


It worked for Anthony Levandowski :)


His story hasn't come to an end yet. He will definitely end up losing more than he gained at the end. Google isn't taking this too lightly. They'll make an example out of him.


> the incumbents -- Intel, Qualcomm and Nvidia -- are massive, and Google, Apple and Amazon are developing their own silicon

I wonder if the death of Moore's Law finally opens the way for entrants.


More importantly it's the death/slowing of Dennard scaling.


With Nvidia pretty much a monopoly in the hardware space for deep learning I think there is certainly a lot of room for disruption. Given that they've got 8 out of the first 10 members of the TPU team there might actually be a bit of potential for this venture to succeed. I for one am happy to see some real competition coming to this market.


A high tech incumbent with a monopoly doesn't scream opportunity to me. This startup is brave AF, but definitely 8/10 from the TPU team have the best shot of anybody.


Let's see who poaches these engineers next


Where do I apply?




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: