AIMunchkin

Well if they're like Bose, you'll be fine as long as you don't install their helpful app.

I could conceive of a high-fallutin' fidget cube for your mind constructed from this technology.

I could also see it used for both a lie detector and a biofeedback tool to defeat its use as such.

But what really comes to mind is the final scene in Firefox wherein Mitchell Gant has to subvocalize in Russian to fire the rearward missile on the fancy Russian jet he's stolen.

If he's from China (and I'm assuming so), I'm not so surprised. It's a status thing culturally. I've had my ideas at work dismissed on the basis of my not attending an Ivy League school by the sort that spouts concepts like this.

A more amusing instance of this outlook at work: http://www.cracked.com/personal-experiences-1940-china-one-p...

Ergo we should immediately arrest all gunmakers because it's hard to think of another use for guns besides killing things or practicing the art of killing things, no?

GP100 supports FP16 FMAD GP102 supports INT16 and INT8 MAD with 32-bit accumulation

Overall, not impressed with Stratix 10. It won't be cost effective, it's not much more power-efficient, and Volta will likely leapfrog it across the board within a year.

Wasn't this thing supposed to sample in late 2014? Back then it would have been a gamechanger at any price. Now, 1080Ti for $700 beats it across the board in throughput/$. NVIDIA's confusing messaging about using consumer versus professional HW is about the only thing that might make it viable for deep learning. Although I note the absence of training perf numbers here, just (apparently) inference.

My cat stayed by my side to her last day, whereupon she decided to hide in the box I had prepared for her body a week earlier.

She had wasted away to 2/3 her normal weight and was having a horrible time breathing, it was time, and it sucked, even recalling it sucks.

Same thing happened to my cat, I took it to the vet to have a bladder stone removed, and I left with a 2 month terminal cancer diagnosis for her. At least she got a month of life without the bladder stone before she wasted away.

Stupid stoic cats indeed. The only way I could tell she had the bladder stone was when she started peeing outside the litter box. Miss her to this day.

"so any one client 'firing' you is suddenly not a huge thing since your income is distributed across multiple clients"

For whom you have to spend roughly 1/3 of your time on collection because companies don't get rich writing a lot of checks. If it fits your psyche, great. Not everyone is like you.

"But for you, I like the cut of your jib, $100M."

Sure, that's possibly true. But if the ensemble of governments planetwide are mostly doing worse than the US(1), then you need to consider it might not be humanly possible to do much better even if it's obvious how to do better theoretically.

But, by all means, go get some billionaire to buy you an island or build one for you to test out your ideal black spherical cow ideas for government. The US itself was once a crazy experiment in cutting edge governing theory itself.

(1) We're not the best, but we are in the top 10% here, no? But I note that everyone who beats us has much smaller populations to satisfy. https://en.wikipedia.org/wiki/World_Happiness_Report

Well yes it is actually, by collecting more data, you are doing a better job marginalizing the distribution of government corruption and waste across the planet. And that can make a seemingly bad thing turn out to actually be relatively good and vice versa.

Or, to quote some famous dead dude: "The World Is a Book and Those Who Do Not Travel Read Only One Page" - St Augustine

Or, to quote Anthony Bourdain himself: “Travel changes you. As you move through this life and this world you change things slightly, you leave marks behind, however small. And in return, life—and travel—leaves marks on you.”

So what would happen if someone created a youtube account called "NotUCBerkeley" and reuploaded all these videos with their playlists intact? Asking for a friend of course!

As someone who has early adopted technology since childhood, I'm surprisingly contrarian on this. And it's not the tech, it's the state of society and industry that's making me feel this way.

For I want said technology to all but disappear Avatar-style except when I need it(1). Along the way I don't want craptastic screendoor VR, I want BSG-reboot Cylon VR. I don't see that coming anytime soon.

I don't want to be bombarded by the outputs of conv nets, SVMs, and other assorted RainMan-level ML models that are occasionally helpful but mostly just distracting factoid spam (see Google Now for a perfect example of this).

I'd love a self-driving car (for real), but I'd love a life where I didn't have to drive everywhere even more. I don't need AI for that, I just need to move to a city. And honestly, driving my sportscar is fun when it's on nearly abandoned mountain roads and highways (see craptastic VR bit about why VR is not a good surrogate). Why would I want to give that up to run with the cool kids(tm)?

I'd even love a brain computer interface, in fact MIT Neuroscience turned me down for admission despite my GPA and GRE scores specifically for saying I wanted to work on this a couple decades ago, but the mobile web is godawful enough already without giving silicon demons like Google and Facebook a direct feed to my brain. I don't trust industry 1 QBit here.

Finally, to quote RadioHead, "I want a perfect body, I want a perfect soul(tm)." If that makes me a creep, I'm OK with that. If AI (and tech in general) doesn't improve me or my life, I don't want it around me anymore and we're just a few years into this.

1. I work in AI. And I'm trying to make it work behind the scenes rather than in your face. To that end, I focus on pull, and I despise any sort of push short of protecting me from harm or keeping me on my schedule.

Well, why don't you learn a concurrent language like Go, CUDA, or OpenCL and do something about that?

From my vantage point, I remain amazed that people have fled concurrent programming in an age where the hardware for doing so has not only thrived, but which may also be the only path forward from here.

These days, I've noticed hip phone menu systems hang up on you when you try to destroy America chanting this obvious codeword to commence the attack.

It's not a best programming language fight. In my experiences in the industry, an enormous of amount of technical debt and operational inefficiency is accrued when someone ignorant of how machines and processors actually work (SIMD, cache, pipelines, threading, etc) is in a leadership position to dictate the toolset for solving problems.

This wasn't a noticeable issue until about a decade ago. But it is now and it continues to get worse IMO. The "programming language" bit is just one of its symptoms when the root cause is ignorance of practical computer architecture.

That said, the mentality of throwing all big data problems at Hadoop clusters with 4 year-old GPUs and flaky 10 gB interconnect (many of which could be solved faster on one.big.modern.and.cheaper.machine(tm)) is working wonders for my Amazon stock so maybe I should just shut up and get rich?

IMO (to be fair, some) CS people became engineering bottlenecks the day the universities switched out teaching C/C++ for Java and Python (IMO the Why Not Zoidberg? of programming languages). Those who learned C/C++ anyway became my heroes.

I have sat through too many presentations obsessing on HW-level perf/W especially w/r to Deep Learning ASIC wannabes. Just writing one's code in C/C++ (and doing it well) guarantees at least a 2x improvement over Java and a 10-100x improvement over Python. I won't even bring up the computational coup that is CUDA.

But hey, let's base a mobile phone OS on Java and block low-level access to its GPU, that's a fantastic idea, right?

See also many experiences with data scientist and CS primadonnas dismissing low-level coding as "ops." I liken this to the Eloi dismissing the Morlocks as "the help."

How about unlocking access to ImageNet? Unless one has a .edu account, its overlords seem to ignore requests to access it. Mind you, it's relatively easy to social engineer access to it, but why should this be necessary? OpenAI and Google have both knocked it out of the park with easy access to datasets and examples.

But sadly, IMO at the amateur-level, TensorFlow considered harmful. I have repeatedly observed novices blow the thing up by starting from one of many of its amazing and fantastic teaching examples. It's not a question of the TensowFlow API, but rather of the engineering quality of its underlying engine, which kind of sucks. Nothing ruins an enthusiastic data scientist's day like a cryptic seg fault for no apparent reason whatsoever.

And I know they're working on it, but fer cryin' out loud, the API is great, and Google has the bottomless pockets to do a lot better than this. It's been over a year and I still see people throw their hands up in frustration trying to make use of the thing. Of course, Google has never been a customer-driven company, but if we don't want an AI Fall, methinks this needs to be fixed.

I'm going to post a contrarian viewpoint that unless your product is self-contained and it can continue on without you if you are hit by a truck or even acqui-hired for top $, I will avoid you.

Too many pay apps and even games with DLC have gone poof in recent years because of the latter situation or because the cost of maintaining the DLC no longer justified itself.

If they like your game, they'll turn it back on.

And it's nice to espouse principles like yours, really, it's nice, but the spammers don't care about your principles and they're making everything suck. Perhaps Google should only do this for sites they haven't validated as a good net citizen?

As someone who has worked at several of the listed companies in that plot, that's significantly more accurate (IMO of course) than either Glassdoor or the Salary data.

100% agreed, given that newer cars now have 4G connectivity, one ought to be able to subscribe for a modest fee to download it.

True, but more likely I suspect is that in 2018 and beyond they'll offer a navigation option. Then again, they still don't offer blind spot indicators on the corvette so who knows?

As for me, I just bought a car this year, and while the next one will be electric, I'm pretty happy with the current set of wheels.

In my case, I'm still quite happy with the Nav system in my wife's 2009 Rav 4 actually. It works as intended to this day. It's missing some fanciness, but I'm just not about the fancy.

Au contraire, I much prefer my 2016 car's built-in navigation system with consistent performance and features to the ADHD ever-changing, signal-dependent, random notifiers asking stupid questions, craptastic voice input, arbitrarily failing navigation apps of the two major mobile providers.

I don't WANT to update my navigation system. That's exactly what I HATE about Google Maps etc. In fact, I have gone so far as to opt out of updating my car's system to enable Android Auto after I tried it out in a rental and found it to be an epic bowl of buggy fail that nearly got me in an accident.

And of course, that didn't stop the Android team from breaking basic Bluetooth support in Android 7.0.1 on my Nexus 5x because they're go-getters in that department.

My car is a useful device that I need to count on daily, not a @#$%ing mobile app like Candy Crush.

And yet no gun camera footage or even pictures...

Funny that...

Old timer rant:

IMO Machine learning mostly doesn't work (yet) with a couple exceptions where tremendous amounts of energy and talent have made that happen. For example, image processing with conv nets is really cool, but the data sets have been "dogs all the way down" until very recently. And for the past few years, just getting new data and tuning AlexNet on a bunch more categories was an instant $30-$50M acqui-hire. Beyond a few categories, its output amuses and annoys me roughly equally.

But the real problem with ML algorithms IMO is that they cannot be deployed effectively as black boxes yet. The algorithms still require insanely finicky human tuning and parameter optimization to get a useful result out of any de novo data set. And such results frequently don't reproduce when the underlying code isn't given away on github. Finally, since the talent that can do that is literally worth more than its weight in gold in acqui-hire lucky bucks, it doesn't seem like there's a solution anytime soon.

Voice input? You gotta be kidding me. IMO it works just well enough to enter the uncanny valley level of deceiving the user into trusting it and then fails sufficiently often to trigger unending rage. Baidu's TypeTalk is a bit better than the godawful default Google Keyboard though so maybe there's hope.

GPUs? Yep, NVIDIA was a decade ahead of everyone by optimizing strong-scaling over weak-scaling (Sorry Intel, you suck here. AMD? Get in the ring, you'll do better than you think). Chance favored the prepared processor here when Deep Learning exploded. But now NVIDIA is betting the entire farm on it, and betting the entire farm on anything IMO is a bad idea. A $40B+ market is more than enough to summon a competent competitor into existence (But seriously Intel, you need an intervention at this point IMO).

Machines with lots of CPUs: Well, um, I really really wish they had better single-core CPU performance because that ties in with working with GPUs. Sadly, I've seen sub-$500 consumer CPUs destroy $5000+ Xeon CPUs as GPU managers because of this, sigh.

Container systems? Oh god make it stop. IMO they mostly (try to) solve a wacky dependency problem that should never have been allowed to exist in the first place.

The web: getting crappier and slower by the day. IMO because the frameworks are increasingly abstracting the underlying dataflow which just gets more and more inefficient. Also, down with autoplay anything. Just make it stop.

My favorite Edison quote:

"Opportunity is missed by most people because it is dressed in overalls and looks like work."

One of the amazingly dumb things about my Nexus 6 is that when it's listening for my voice it still plays incoming SMS message tones that it then takes in as part of speech recognition and messes it up. Why? Really, just why?

I end up having to correct it with typing anyway.