I was employee #1 at a startup. Right out of the gate our medical was better than Google’s. Having been exposed to the financial side of things, the cost difference between shitty and decent insurance for employees was inconsequential in the overall cost of doing business.
+1. I want a book that walks me through a real, well designed, functional (if you pardon the pun), useful piece of software, end to end. All these FP books are too enamored with language features, instead of showing you how a programmer proficient in the language would solve practical problems in it.
I haven't looked into it myself, but it could be that due to all this massaging you'd lose more on throughput than you gain on memory use. It's similar to doing 8 bit quantized stuff on general purpose CPUs. It's very hard to make it any faster than float32 due to all the futzing that has to be done before and after the (rather small) computation. In comparison, no futzing at all is needed for float32: load 4/8/16 things at a time (depending on the ISA), do your stuff, store 4/8/16 things at a time. Simple.
Ours was very large. In fact, I'd say, a 2-person do-nothing prime would be a vast improvement over what we had, especially if they did not have any illusions about their own level of competence. To see the situation in the positive light, however, I have vastly improved my project management skills. Making _anything_ stick on the other end was a non-trivial endeavor.
Been there, done that. I feel much worse about paying my federal taxes now. Federal contractor (to whom we were a subcontractor) spent at least five times what they should have, and very little of it trickled down to us. They also were a “broken telephone” between us and the end customer, so we didn’t get much user feedback. Any call (not _video_ call mind you, a phone call) with them had a minimum of 10 people, and after the call everyone on their end would have immediate amnesia. 1/10 would not recommend.
>> Why not hire the best no matter where they come from?
Here's what H1-B actually works if you hire "the best": you hire people who are way overqualified for the job, and pay them peanuts while they are on H1-B. "But H1-B requires that they are paid no less than US workers" you might reasonably object. There's a loophole for that: just hire them at a way lower job ladder level than an American would consider adequate, and don't promote them as much. It's not like they have much of a choice anyway: they can either work here or GTFO to their country of origin within 2 weeks.
I liked Trump's proposal that only the highest paid positions get H1-B visas. That way economic incentives essentially turn H1-B into an auction for the truly great people, and make life easier for fresh US grads. I'm not sure what happened to that proposal. It was probably smothered in the crib by FANG lobbyists.
Full disclosure: I'm a former H1-B, now a naturalized US citizen.
IIRC all recent NVIDIA GPUs support fp16, it's just that the perf of fp16 is severely hampered on "consumer" grade hardware, so the memory/perf tradeoff is not viable. I mean, on the one hand I can see why that is: fp16 is not terribly useful in games. But on the other, it's probably nearly the exact same die with a few things disabled here and there to differentiate it from $7K Tesla GPUs, which as an engineer I find super tacky, much like MS deliberately disabling features in Windows Home that don't really need to be disabled.
Whether it's substantially better for deep learning depends on whether they chose to neuter fp16 like they did in all other consumer GPUs. The main bottleneck is not actually the size of the model per se (sizes of the models have been mostly going _down_ as of late), it's the size of the batch during training. Both batch size and model size determine the size of the main memory hog: activations that you need to keep to do training. You don't need to keep activations to do inference: you can discard them after they're fed into the next layer. So it's not entirely accurate to say that this is useless for DL. What could make it totally kickass for AI would be a proper, non-neutered implementation of fp16, similar to what you'd find in P100, V100 and (surprisingly) Jetson TX2, which would effectively double the possible batch size, which in turn leads to quicker convergence. It could also halve the model size as an added bonus. This is NVIDIA, however, so I would bet good money fp16 is still neutered in "consumer" SKUs.
Exciting to hear. .NET is truly a diamond in the rough on the Linux side, its only major problem seems to be catch-22 in the sense that few people use it there, so the ecosystem is slow to come in.
Bing folks, are you running this on Linux? Common problem a few years ago was that perf of the Linux version of Core sucked compared to Windows. The only way to fix that is through serious dogfooding. I’m wondering to what extent Microsoft is committed to such dogfood on Linux.
Can confirm, suffered from RSI as well. I think most doctors are unaware of the benefits. If they were, they wouldn’t prescribe the bullshit they’re currently prescribing. I’m not saying this will help everybody, but it’s 100% worth a try before trying anything more “medical”.
You need to do compound exercises with a barbell, namely low back squat and deadlift. These target the so called posterior chain, which is all those crucial muscles you don’t see in the mirror. Mindlessly “working out” is not going to target all that to a sufficient extent.
I was where you are now, started lifting and my back pain was completely gone within 2 months. I have been totally free of back pain for over 5 years now, going from not being able to lift a 60lbs barbell without pain to deadlifting 500lbs.
5% collateral casualty rate, while regrettable, is very, very low, comparatively speaking. And yes, I was discussing this in context of a purely theoretical invasion of Taiwan. That's the whole point.
It will: https://www.cnbc.com/2018/08/01/trump-clears-way-for-health-.... Downvote brigade has arrived, so I can't respond. This is not what he was talking about, but you gotta start somewhere. I'm on an Obamacare plan myself, and I agree that it's a "total disaster". $13k+/yr for barely any coverage at all.
We’ll see about that in the next couple of years. This was one of the more prominent campaign promises by Trump, and the dude has a habit of actually following through on that, whether you like it or not.
Yep. I take blood pressure medication. Cost without insurance: $5.70/mo. Cost with insurance: $10 co/pay. I know this because my insurance lapsed once for a month. Same exact pills, same manufacturer even.
As the past 60 years of US Military action have made abundantly clear, even for the best military on the planet by quite a margin no war is “easy” to win. Now, granted, China won’t agonize about collateral damage the way we do, but even then it’s not “easy”, especially if you plan to take permanent control of the territory afterwards. For that to happen, the overwhelming majority of the country has to _want_ such control, and the remaining people have to not be too heavily invested in the idea of independence. It’s pretty much the opposite of what you see in Taiwan.
I think one’s self worth is intrinsically linked to acceptance, with or without Facebook, especially in younger people. FB (and others like it) just exploited the weakness.
Keep trying, and don’t take it to your heart. The whole interview circus is stupid and everybody knows it. The only people who think they’re the smartest around are the ones who never worked with folks who are truly, freakishly gifted. I’ve been lucky in that regard, so I no longer have any illusions about my intellectual prowess. Over time you will see that intellect isn’t everything, nor does it guarantee success. Stay in the industry, change jobs every 2 years, put yourself in the path of serendipity, and do a good job. Mobility will come eventually.
>> don't hinge your happiness on something as random as an interview decision
This is easier for senior folks. At some point you just know the rejection was a failure of their interview process and not your ineptitude. I have failed several interviews at different companies where I was one of the best people _in the world_ for the job due to my domain expertise. It's a roll of the dice for everyone, but earlier in my career I did not take rejection well. Now? Honeybadger don't care.
Sure, but you did not narrow down the audience. People as a whole, it would seem, slightly prefer checkout counter magazines as opposed to NYT. Likewise, I'm not sure bite-sized bullshit regurgitation services like Twitter/Mastodon are even useful in the first place, let alone benefit from refined moderation.
I’ve also read that concrete absorbs CO2, in significant enough quantities that an isolated ecosystem experiment had to be aborted because plants weren’t getting enough of it.
>> You know the reason why people prefer reading
>> the New York Times instead of checkout-counter
>> tabloids?
I'm not actually sure that's true. Upon a cursory check, People Magazine has a slightly greater circulation than NYT. Heck, a non-trivial number people on YouTube have greater circulation than NYT or mainstream TV News channels and shows, while filming their stuff in the living room of their apartment.