varelse

193 ms·

451.
by varelse 8y ago
See lecture 15 of cs231n 2017: https://www.youtube.com/watch?v=vT1JzLTH4G4&list=PLC1qU-LWwr... Further, while all these tricks are amazing, they are not currently of much use to fungible data scientists at FAANG and in
452.
by varelse 8y ago
You're right, it was Steve Ballmer who nearly vanquished Microsoft at a time when Google was the company to work for in tech and kept doing amazing things. At least IMO. Unfortunately, by the time of my brief stint at Google, the pl
453.
by varelse 8y ago
Yep, and in doing so, you just traded "feature engineering" for graph design and data prep, no? And that's my response to these sorts. And their usual response to me is to grumble that I don't know what I'm doing.
454.
by varelse 8y ago
I'm 100% convinced it can block 99% of all terrorist content that hasn't been effectively SEOed to get around their filters because that's just memorizing attributes of the training set data. Unfortunately, the world isn&#x2
455.
by varelse 8y ago
Duplex was impressive but cheap street magic: https://medium.com/@Michael_Spencer/google-duplex-demo-witch... Microsoft OTOH quietly shipped the equivalent in China last month: https://www.theverge.com/
456.
by varelse 8y ago
I'd clarify that there is a specific delusion that any data scientist straight out of some sort of online degree program can go toe to toe with the likes of Andrej Karpathy or David Silver with the power of "teh durp lurnins'
457.
by varelse 8y ago
Yep, they're a 3D reconstruction of points from a 4D space a (massively simplifying the fact that the 2D video frames themselves are high-dimensional data) and that's just the video. Bring a 3D camera into the game and the underl
458.
by varelse 9y ago
Indeed, he is killing his darling before someone else can. No interest in working there personally because I despise both PHP and working in an open office, but I don't think the Zuckerberg brand of fine social content is going away an
459.
by varelse 9y ago
Floating point math itself is commutative, but it's not associative as it is implemented in current processors so you've already lost associativity from the get-go.
460.
by varelse 9y ago
I can think of no better way to demonstrate that you are right than to succeed at building your own business. OTOH if the business fails IMO you ought consider that you need someone objective to review your perspective on and approach to b
461.
by varelse 9y ago
Well, either he's pocketing a large fraction of that or he's up to his eyeballs in debt looking at his new boat and his fancy new truck. Methinks you're overanalyzing it.
462.
by varelse 9y ago
My doctorate is not in CS, Math or an engineering field, but I did (somewhat unknowingly) apply ML/statistics to my thesis, which was (more or less) generative adversarial search 20+ years ago. Everything I did then turned out to be st
463.
by varelse 9y ago
My plumber charges me $80/hr in the bay area. Now that's ~50% above the poverty line, but it's >2x the median wage around me.
464.
by varelse 9y ago
Yeah that bubble will burst when AI/ML replaces us or America* creates a surplus of competent engineers. Don't hold your breath for that. I made top $$$ (on a relative scale) in college tutoring aspiring computer scientists and
465.
by varelse 9y ago
Have you tried to get a decent plumber lately? My plumber just bought a fancy new truck and boat. Previously he traded favors with my contractor to remodel both their homes on the cheap.
466.
by varelse 9y ago
In my case, I turned a career in driver and low-level algorithm coding into a career in AI in <2 years emphasizing the ability to implement AI algorithms in C/C++ and CUDA. That said, when Tensorflow Monkeys can make $300K right ou
467.
by varelse 9y ago
Well then look busy!
468.
by varelse 9y ago
I am amazed that rather than see this as opportunity for racking up top $$$ for at least the next 3-10 years by learning Calculus, Probability/Statistics, and Linear Algebra on the way to whatever degree one seeks, there is so much c
469.
by varelse 9y ago
The new driver EULA clause banning the use of GeForce GPUs in "datacenters" except for blockchain purposes has ended my positive impression of NVDA. To that end, we have ceased the use of the term in favor of "GeForce Houses
470.
by varelse 9y ago
So 3 2nd generation TPUs are ~= 1 Volta class GPU ~= $3 per hour on-demand on AWS: https://aws.amazon.com/ec2/pricing/on-demand/ and ~$1 (75 cents at the moment with p3.8xlarge and its 4 GPUs) in spot: https
471.
by varelse 9y ago
Sure, but the whole point of the above idea is to compare our 20W computers to what we can build that eats 20W. And don't give Silicon Valley ideas about disrupting the lucrative Mechanical Turk ecosystem by scaling it up with ideas b
472.
by varelse 9y ago
Sure, why not, but then how do we compare to a human?
473.
by varelse 9y ago
I think we need to start capping total electricity and total $$$. I'd love to see AlphaZero 20W pitted against that other 20W supercomputer. When humans fall to that, be afraid(tm). I'll even be charitable in order to simulate
474.
by varelse 9y ago
That's not clear, each (second generation) TPU is 45 FP16ish unspecific TFLOPs. A single board consists of 4 TPUs at 180 TOPs total. This is similar to the Dual P100 NVLINKed Quadro which is an absolutely killer HPC/DL card. I
475.
by varelse 9y ago
Well, given the Trump tax plan has passed both houses, I expect it to affect any state with high taxes, be they income (California/NY) or property (Texas). Who needs a double-taxed carrying cost like this? House prices will likely dr
476.
by varelse 9y ago
NVIDIA driver stability as of CUDA 9 or so has been craptastic IMO. I am seeing all sorts of strange slowdowns and app crashes that just weren't there pre-CUDA 9. I would assume this is related to changes made for Volta support in so
477.
by varelse 9y ago
As someone who spent a (wasted) decade in academia finding out a lot of published research was faked, keep your academic hands off of my business models. And keep your government hands off of my algorithms. FB stupidity and greed aside, I
478.
by varelse 9y ago
At the risk of falling prey to "WhatAboutism" it is hard for me to be concerned here when we give such free reign to credit agencies and other traditional corporate entities who can't be arsed to even keep their data secure.
479.
by varelse 9y ago
Tell that to my bank account. We live in a world where most current and future performance benefits come from understanding how to program both parallel and distributed algorithms in an age where almost everyone is either a single-threaded
480.
by varelse 9y ago
See also the crazy cash they can make building distributed machine learning/big data systems at next to zero risk...

More ›