|
Carlos E. Perez š§¢
@
IntuitMachine
Washington, DC
|
|
Follows you - Deep Learning with Complex Adaptive Systems
deeplearningplaybook.com
medium.com/intuitionmachiā¦
#deeplearning #ai #machinelearning #nlp
|
|
|
19.223
Tweetovi
|
2.416
Pratim
|
4.532
Osobe koje vas prate
|
| Tweetovi |
|
Carlos E. Perez š§¢
@IntuitMachine
|
18 h |
|
43%. Not bad returns for investing in AI firms in 2019. wsj.com/articles/tech-⦠Anyone want to send me some spare change?
|
||
|
|
||
| Carlos E. Perez š§¢ proslijedio/la je tweet | ||
|
Riccardo Di Sipio
@rdisipio
|
3. velj |
|
Remembering @peter_wittek pic.twitter.com/W4DG4XBUN8
|
||
|
|
||
| Carlos E. Perez š§¢ proslijedio/la je tweet | ||
|
Amir Mizroch
@Amirmizroch
|
21 h |
|
āA lot of VCs look more like private equity companies that do not want to lose any money so they end up backing dog-walking apps rather than quantum computing.ā ft.com/content/34a562⦠via @financialtimes
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
19 h |
|
Slides?
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
19 h |
|
Thanks! Let me which threads you thought had interesting insight!
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
19 h |
|
Yes, Cargo Cult science is science by mimicry but biological cognition initially involves mimicry before there is comprehension. twitter.com/fchollet/statuā¦
|
||
|
|
||
| Carlos E. Perez š§¢ proslijedio/la je tweet | ||
|
Jason H. Moore, PhD
@moorejh
|
3. velj |
|
A global genetic interaction network maps a wiring diagram of cellular function - Science (2016) ncbi.nlm.nih.gov/pubmed/27708008 #epistasis #edge20
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
Life and cognition are driven by far from equilibrium processes.
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
Time to dispense with that fiction that the market is efficient. Certain stocks go parabolic as long as there are shorts to squeeze. twitter.com/benedictevans/ā¦
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
Fascinating reframing of biology by Eric Smith: youtube.com/watch?v=gSZJ22ā¦
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
Allostatic regulation? References?
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
So if you are one of those startups that want to take up this strategy, then do let me know!
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
Every company that is working on Deep Learning solutions do not want to be bothered with the plumbing. Yet the hardware vendors are expected to deliver on the plumbing. But hardware folk don't know software, so they underestimate the complexity and disappoint on the delivery.
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
The money is in recognizing that there's a massive shortage of people who know how to develop this low-level software. Every hardware firm thinks they can get away without software. They are entirely mistaken and will fail because of this lack of insight.
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
The money isn't in specialized DL hardware that nobody can program because of lack of software. I've said this in my 2018 predictions: medium.com/intuitionmachiā¦
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
Yes, find a VC with enough smarts to understand this play and bankroll it.
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
The play is very simple. Find a boat load of young engineers who love coding close to the metal. Also find some serious compiler optimization folks. House them in an isolated location for 6 months. Give product out for free. Cash-out with an acquisition in one year.
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
Nervana's technical chops were demonstrated by doing just this. They reverse-engineered NVidia's instruction set and created hand-tuned high-performance code. But they got bought out by Intel, to do something else entirely.
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
If you want to startup a deep learning (DL) company that is guaranteed to be bought out then focus on optimizing low-level assembly code for DL hardware. This stuff is given away for free but very few have the expertise. It's damn hard work but somebody has got to do it! #ai
|
||
|
|
||
|
Carlos E. Perez š§¢
@IntuitMachine
|
3. velj |
|
Nobody can seem to figure out why Intel axed Nervana in favor of Habana. My hunch it has to do with the complexity of Nervana hardware and software. Customers only care about raw horsepower and not complex adaptive designs. fuse.wikichip.org/news/3270/inte⦠.
|
||
|
|
||