1,911 karma · joined August 28, 2018
I think the most detail I can give is general class of work it engaged in. Think of MRIs, sonar, radar, ultrasound, back-scatter x-ray, LIDAR, CAT scans, etc. That's the general sort of technology it worked on.
That should give enough detail for the purposes of this discussion...
Best company I worked at -- small startup -- would always do 2-3 R&D initiatives in parallel. One would be conservative (guaranteed results). One would be super-high-risk (huge payoff if it works). Sometimes there would be one or two more. It was all a big hedge. Some panned out, some didn't, and when we hit the market, our technology was like science fiction. Competition didn't know what hit them.
That would run a fair bit of coin, but fairly little on the national scale.
(Footnote: Not a software company)
> I’ll start with the bad news: Nobody will be making an mRNA vaccine in their garage any time soon.
... We're not there in time for COVID19, but technologically, I think we're less far away than this article presumes. Cost of both sequencing and synthesis has been dropping like a rock, albeit not very smoothly. I can get DNA synthesized for less than what I was paying for amateur PCBs as an undergrad.
A lot of the other complexity, I suspect, is specific to getting this out in <1 year. It's worth remembering the earliest vaccines took no technology beyond what I have in my garage. With a little more time and patience, I'm sure some of the other complexity will go down. We can't spare time and patience when we're bleeding billions of dollars from our economy and thousands of lives each day, but....
> How is that subtext not correct here?
The grammar is a little bit hard to parse in the original article, and I don't know the cultural context, so I don't so much have AN interpretation as a probabilistic cloud of possible interpretations.
The thing I would remember when reading anything like this is how much language means different things in different cultures:
1) Words like "racism," "white supremacy," "gaslighting," and "abuse" get tossed around a lot in younger, ultra-liberal circles. If I call someone conservative in their fifties a white supremacist, I won't be welcome back ever again. If I'm talking to a liberal 20-something, we're both expected to accept our white supremacy.
2) Positive / negative language varies between cultures. If an Eastern European says something is mediocre, they mean it's typical. If an American says the same, they mean it's bad. Some cultures use language in ways which are hyper-positive, and others, in ways which are hyper-negative. Acceptable degree of exaggeration varies too.
I recently had a "friendly" lawyer for an adversarial party provide me with some "friendly" legal advice (which was specifically that I had no case, and would be out both sides' legal fees if I went to court, and so on). They were lying. I knew that, and they knew that. This was an attorney for a megacorp, where I filed an arbitration complaint as a typical consumer (so they probably didn't know that I knew that they were lying).
With filters for how the author might be using language -- well within the norms of some cultures I've interacted with -- I didn't see this as going far beyond "If an opposing counsel tells you that you have no legal case, they might be lying to you."
This sounds cynical, but it's not nefarious at all:
1) As a manager (at various points in my career), I have better things to focus my time on than sick policies, vacations, sick leave, or similar. I'd rather focus on mentorship, growth, strategy, and technology. Outsourcing the menial stuff to someone else means I can manage people better. Plus, they're more likely to do it right. If I have a 16 year old intern who is a resident of Ohio working remotely whose parents are on an H1B visa who wants to do an unpaid work-sponsored training program beyond their 40 hour work week...
2) As a manager, I have no background in how to resolve legal issues. If an employee engaged in improper behavior (harassment, theft, etc.), the last thing you want are amateurs handling this. It's a relief to offload that onto professionals, both for the manager (who doesn't want the stress or liability) and the company (who doesn't want the liability).
... and so on.
HR is the company's agent. It's not your friend. It's job isn't fairness, anonymity, or justice, but to protect the company. It's still a necessary function and having it there is not evil, so long as employees don't go in with the expectation HR is some impartial judge. It's not. A lot of kids come in and treat it that way, and get into a lot of trouble.
* Executive ethics focus on maximizing shareholder value, while individual incentives, on personal returns. If I can pollute a lake for an extra buck to my stock price, I'm expected to do this.
* HR ethics focus on protecting the company. If they can lie to you as an employee to minimize legal risk, that's what's gonna happen.
* Legal ethics focusing on protecting your client (no matter how evil), while practice focuses on maximizing billable hours.
... and so on. It's not good or bad. It just is. As a programmer, you'll be off-putting to people from those groups for reasons just as valid or invalid.
And generalizations are often accurate. If you don't show a certain modesty and humility in gift-receiving in China, you'll be an outcast. If you don't show a certain opulence and over-the-top gratitude in some parts of the Middle East and show the same, you'll be an outcast just as much.
Guess what happens when the two cultures come together?
People are viewed as jerks.
A little cross-cultural training goes a long ways to avoid that. I'm sorry you don't like the tone, but I viewed this as a fair guide for programmers managing corporate settings running into those barriers. This article appears to do a fine job explaining HR's role to people coming from e.g. CS undergrad degrees which will prevent them from getting hurt. It's no better or worse than a guide for e.g. American women as to what to expect if they marry someone from Saudi culture.
As a footnote, quotation marks suggest you're quoting someone. Your claimed quote doesn't exist in the source article. Your point would be stronger if you commented on what the author wrote than your (somewhat inaccurate) read-between-the-lines. That's another place cultures differ a lot: how things are implied and subtexts. People misread subtexts, which I think you did here.
What I care about is good intent and reasonable measures to support me for the lifetime of the product. In most cases, I can't get replacement parts because manufacturers discontinue them after a few years as a business choice, not because they have to. Most manufacturers rotate product models every year, and I presume it's expensive and not good business to stock parts for models from a decade ago. That's what I want to avoid. I'd like a manufacturer that will continue supporting me as long as reasonable.
What I do now is I look at manufacturer's web pages, and see the oldest products they still support. It's pathetically short.
My last vacuum lasted 15 years before replacement parts ceased to exist. I've been looking at upscale Dyson vacuums, and the oldest ones with replacement parts are a decade old from date-of-introduction (less from date-of-purchase, of course). For a $800 vacuum, that means you're looking at an annualized cost of $80-$160 before you're dumped on the curb.
Of course, there are companies which simply don't offer replacement parts. If a roller or belt fails, the vacuum is disposable. The TCO seems similar, since those don't run $800, though, and a $150 vacuum will probably last a couple of years before something breaks.
What I'd really like is a web site which compiles this sort of information.
The government, at some point, decided to regulate utilities. It certainly didn't create them or their monopolies. Most have a natural monopoly (https://en.wikipedia.org/wiki/Natural_monopoly).
It was a good idea, at the time, and it's just helpful to update our list of what constitutes a utility.
My lesson here is that the Internet is essential infrastructure, and once companies reach a certain size (Google, Facebook, AWS), they should be regulated similarly to public utilities.
An electric company shouldn't be able to cut off electricity to a crack house or brothel unless law enforcement first asks for it. Law enforcement should have due process appropriate to the situation; a terrorist attack should be cut off right now. A company making pirated copies of Windows should have their day in court. Your propaganda-believing gun-toting drunk crazy uncle should probably keep their electricity (which isn't to say he shouldn't be handled somehow, but perhaps committed to a medical facility).
I don't feel good or bad about them being taken off-line, since I don't actually know what they hosted. I do feel bad about that decision being made by corporations behind closed doors.
In practice, you're not finding a bug like this every week.
The bug bounty programs were originally intended to give a white hat market alternative to the black hat and gray hat markets. They don't do that. If I find a bug, and I want profit, I'm much better off selling to my government than to Google.
One can only imagine the number of exploits the US, China, Russia, North Korea, etc. have in their cyber-warfare vaults.
Exploits compound. Often, two minor exploits make a major exploit.
There's a gradual process to go from hypothesis to theory to fact, and to get there, you need several types of independent evidence. Correlational evidence is okay as one of those.
A correct statement is that correlation is not proof of causation (no matter how strong).
Types of evidence:
* Correlation
* Theoretical basis / strong hypothesis
* Extrapolation
* Interpolation
* Small-scale well-controlled experiments (lab setting)
* Large-scale less controlled experiments (real-world setting)
* Anecdotes
* ... and so on
You want several of those before you start to believe anything, and some are stronger than others. Correlation isn't fundamentally weaker than most of those, though; all of those carry their own methodological issues. The number of times you can have a large-scale perfectly-controlled preregistered randomized control trial with no confounding effects is exceptionally rare (some medical trials, and a few other settings).
* Proprietary
* Around 4x the price
* With less community
* With worse documentation.
* With some learning pathway based on Dr. Who videos, which you can't preview before buying
* 5x the speed, colored LEDs, an oversized case and wifi.
I like Micro:bits, but part of the appeal is I could buy four of them to start off with, and expect to buy more over time. They're cheap enough to power projects, and they can communicate with each other.
I also really like Makecode with the Micro:bit. It seems a lot nicer than Tynker.
Doesn't really do it for me.
What would do it for me is a Micro:bit-compatible device with a basic touchscreen. One of the limitations of the Micro:bit is limited input. If I have Micro:bits controlling lights, or robots, or whatever, it'd be helpful to have a single "master" controller where I can control things from.
(2) Traceability. git has secure hashes, and things can't change when you're not looking.
My experience is that Google cares deeply about its own security, but not much about the security of its users. This sort of change is reasonable, but completely outside of Google's psyche. Google will
(1) Silently disable Android updates, leaving many running exploitable phones
(2) Hold back security tools for Google Apps without a premium subscription. If your account was compromised, you have no way to do audits to understand what happened without $$$, which leads to many more attackers.
(3) Expires Chromebooks rather quickly. Fortunately, unlike Android, it lets users know, but given the target market, many can't afford to upgrade.
(4) Runs appstores full of malware. When malware is discovered, users have no way to know what it did. They're just notified malware existed.
(5) Doesn't allow any sort of reasonable sandboxing of Android apps. If an app asks for filesystem, maps, and other permissions, you need to agree to run the app. I can't have Android give a dummy location or otherwise
Given that the bulk of Google's business model is built on mass surveillance for advertising, with users-as-statistics, this isn't too surprising, but it's something to be aware of if you use Google.
I firmly believe in civil liability for software companies which ship insecure products. They shouldn't be able to externalize costs like this. Follow good security practices, or your insurance premiums go up.
1) Allow both private and public right-of-action for most laws. If the AG is busy, I should be able to sue. If I can't afford to sue, an AG should be able to take it up on my behalf.
2) Go back to circa 1800 style courts, where you don't need a lawyer to represent you. You both make your case to the judge. Not too much procedure. Perhaps extending small claims court up to $100,000 would do much of the same.
The answer to that might be right now (we couldn't do it before, and we probably can today) or it might be in a decade. But electronics will keep falling in price. 10,000 microphones * $5 per microphone+electronics = $50k.
The advantages of cancelling an entire wavefront go well beyond passenger comfort too. Industrial noise cancelling systems are more about equipment life than about employee comfort. I had laptop screws unscrew on airplanes before, due to vibration. If planes need less maintenance as a result of active vibration reduction throughout the airplane, you'll make up that $50k virtually overnight.
As a footnote, the crazy part here isn't the 10,000 microphones, but the speaker-at-every-seat part. You'd almost certainly want both the microphones and speakers in the skin of the airplane. But that's a story for another day.
And in either case, the design of algorithms depends on much more on context than on whether it's "consumer" or "industrial." In this case, the context looks a lot more like industrial machinery.
1) Assuming system-level model
10,000 microphones * 8 kilosamples per second = 800 megasamples per second.
NVidia Tesla does 100 teraflops. I get about 125,000 FLOPS per sample. That feels adequate to me!
2) One filter per microphone per passenger. We need to divide by 250.
500 FLOPS per sample per passenger. That's more than enough for a very fancy IIR.
Of the two, I think #1 is more likely to work than #2, precisely because you want a coherent model. If you want to adjust your model for someone walking down the aisle (or any kind of system ID), that's a lot easier with 125,000 FLOPS per sample than 500 FLOPS per sample.
Well, let's do the math here. I can't find the data sheet for the weight of an electret microphone cartridge, but let's assume 4 grams. A thousand microphones -- assuming we want 100 the length of the airplane and 10 around -- is around 4 kilos. Add another 2 kilos for computation, and a few more for wiring, and you've added the weight of a piece of luggage.
Now, let's say you want to add 2 inches of mineral wool sound insulation. I'm assuming 100 meters x 10 meter circumference (I have no idea, but the numbers scale the same). You've added (quite literally) half a metric ton to your airplane. And made it 2" thicker, either reducing cabin space or increasing drag.
Plus, sound insulation does very little for low frequencies, which is the majority of airplane noise.
Speakers, located in the seats, if they knew where ears were, could play sound to cancel that.
Decent microphones are $2, and what was expensive were things like microphone preamps, ADC, and the whole data pipeline. From there, we need a shit-ton of computation which really wasn't practical until... big-ass computation systems came out for the rise of ML.
The problem is hard today, but definitely not impossible. I don't think I would have given that same answer a decade ago. An NVidia Titan V brings over 100 teraflops. That's a lot of cycles one can throw at trying to predict sound by my ear from sound at the skin of the airplane in real-time.
Noise-cancelling headphones work the way OP described, and are limited to working in a cavity which is smaller than the wavelength.
There are predictive noise cancelling systems too. These are commonly built into industrial equipment. You might, for example, have a fan or motor which spins at 5000 RPMs, and what do you know, it has a consistent noise profile on each of those rounds. If you cancel that noise profile, you do pretty well. These systems both reduce noise and extend equipment life by reducing vibration. They also work to higher frequencies (in part because speed-of-sound in metals is > 10x that in air, and what matters is wavelength more than frequency).
There are ones which are open-loop used in HVAC systems. If I have a pipe, I can record sound on one end, and play a cancelling waveform on the other end, with an appropriate delay. I haven't followed the field, and when I looked, these were mostly expensive prototype systems (I'm not sure if they ever made it into mainstream use), but they did work. They needed to be calibrated for each HVAC, which made them impractical for most real-world applications.
There are all sorts of ways to do active noise cancellation. Many of these could be implemented on airplanes. Most of these techniques were implemented decades ago, when computation was a lot more expensive. A simple feedback loop is a few opamps, capacitors, and inductors, so pretty cheap to toss into headphones with 1990's-era technology. One could do a lot more with teraflops....
What I've seen:
* A person is tricked into thinking software engineering is like baby intro "get into CS" projects, and by the time they find out it's not for them, it's too late. I know several people who went this route, and range from having gone on to law school/MBA to working retail. The law school/MBA types are extremely successful, with the dual background. But I feel pretty bad for the folks working retail, with crushing engineering school debt.
* Companies have a severe shortage of qualified employees. Under-qualified employees have a severe shortage of jobs. I feel bad for people who can't find jobs.
My own opinions:
1) No, code camps are not sufficient prep.
2) No, application-focused CS undergrad programs are not sufficient prep (where you'd historically learn Java web development, or the current Microsoft stack)
3) Yes, having rigorous background in something, be it machine learning, low-level systems engineering, compilers, or what-not makes you far more effective. Ideally, a few.
4) Yes, freshman weeder courses are important
5) But no, having "highly-selective schools" like MIT or Stanford is not helpful (and no, it's not sour grapes, I went to one of those). Anyone should be able to try a freshman weeder course, rather than going through insanely competitive admissions criteria which correlate poorly with the ability to do this stuff.
I will say familiarity with code varies person-by-person. I'm generally known as at least a 10x developer, since I've created several major tools with name-brand recognition. I'm not great at jumping into other people's code, though. The article would have resonated with me if everyone's brain was wired like mine, but...
... one of the best developers I've worked with can jump into complex codebases in what seems like overnight (a couple of weeks to be literal), and work with them better than the original developers. You toss him a complex system with a bugs, and it will be debugged in a month, where it'd take me a year or two of work to even understand that code. On the other hand, I wouldn't trust him to write a new piece of complex code.
So it seems we're wired differently. I'm better than he is at imagining a new system, foreseeing possible use-cases, designing an architecture that's general in the right ways to allow all conceivable uses, yet constrained in the right ways to still be simple, and then building it quickly. I'm better than him at designing novel algorithms and data structures too. On the other hand, if I have a bug in a third-party library, I'll take literally 10-100x as long to fix it as my friend. And he'll be better at understanding the algorithms and data structures in existing systems (you can throw something complex like Cassendra at him, and he'll understand the design and its implications nearly overnight).
Good numbers are hard to come by since:
1) Much of the wealth inequality in Europe does a nice job of hiding wealth. Old money doesn't talk about wealth. But there are families in Europe with fortunes dating back a half-millennium. A half-millennium of interest, even at the relatively low rates over most of history, is a lot of money.
2) The figures are biased by people with negative wealth (e.g. debt) in ways which are somewhat different from what one would hope to understand.
A model to follow is based on a modernized version of Islam -- a wealth tax.
If you have a candidate from Harvard, it's a mixture of ALDCs and really smart people with some bias for policies like affirmative action. You might not know what you're getting, but there are odds of scoring that genius. Ditto with most other low-quality signals.
And statistically, you do get someone better than a random schmuck off the street. I mean, the processes are noisy, and perhaps 10% of your workforce might make a better CEO, but you have no way of finding those 10%.
(And the noisy signals problem in hiring isn't specific to CEOs).
If I want morbidus-rex.com, and I see a cybersquatter there, I can file a complaint. I pay $100.
You also need to give appropriate tools to allow private enforcement. If I file a complaint, the cybersquatter is required to release documents showing corporate ownership all the way up the chain. Failure to do so results in automatic loss.
Registrar arbitrates, with first $1000 in costs going to complainer, and remaining going to registrant. Typical case, I show that I'm an individual or similar entity, and it's done. If there is a complex corporate scheme to decipher, the party with the complex corporate scheme covers costs.
And if you hold a thousand domains like jkhzxkljhdsjkl12jkl.com, I don't really care.
1) First-come-first-serve for your first handful of domains. Price is nominal (e.g. $1/year).
2) Increasing price with number of domains held. If I hold myname.com, it's $1. If I hold a thousand domain, the last 500 are at $1000/year.
3) Oversight for transfers. If sold, resale value is capped. Historic domains shouldn't be used for spam but perhaps go to archive.org or similar entities.
4) Managed by a government institution or not-for-profit with open meetings, salary caps, public records requests, and other (legally-enforceable) checks-and-balances.
5) Several TLDs, but you can't register in all of them. You can have mybusiness.com or mybusiness.mobi, but not all of the above. If I have the same business name as you (Apple records v. Apple computer), I go to a different TLD.
6) Rules for new TLDs. You can't pull the Coursera.org "We're a not-for-profit" shtick. Old domains grandfathered in.
7) Oversight/governing board with members from the EFF, FSF, OSF, and so on.
Open auction, with price going as high as it wants, right down to zero.
Special categories (e.g. multiple co-morbidities, medical workers) cut to the front, and pay market rate.
Let businesses foot much of the bill. If I need to pay $1000 to reopen my factory, it might be a good investment. Free market will allocate resources.
If a CEO causes 1000 employees to be 1% more efficient, that creates the same value as 1000 employees each being 1% more efficient. Ergo, if I'm an investor, if I believe a CEO will be slightly better than another CEO, it makes sense to pay a lot more.
If I have a market of 7 billion people in the world, and my software saves everyone a dollar, I've created $7E9 of value in the world, and probably monetized $7E7-$7E8 of value for myself. 40 years ago, my market was the US -- roughly 300 million people. With globalization, it's 7 billion.
If I'm making a search engine or webmail client, it's winner-takes-all. If I can hire two engineers, and I believe (statistically, per my lousy interview process) that one is 1% better than the other, I'll drop a shit-ton of money to hire them. See Mythical Man Month for why hiring an extra person with the "saved" money doesn't have the same effect.
And so on...
Economies of scale, winner-takes-all-markets, and zero marginal cost goods like IP fundamentally push salaries at the top up, even with small differences in performance, measured with noisy tools. I've worked with many CEOs. Someone who makes $10M isn't necessarily brighter than someone who makes $100k, but the demand-side economics still work out that way. The $10M salary often comes down to signalling; they've managed a big organization before, they have a nice elite university brand stamp, or they felt better the day of the interview. It STILL makes sense. My expected returns are STILL higher.
Where things break down a little bit is corruption and misaligned incentives -- lots of places, if you pay more, you get less -- but I think that's under-acknowledged. Things also break down is Simpson's Paradox. Globalization means income inequality WITHIN countries is getting worse, but worldwide, is getting better.
Nikon, if it were to survive, would need to move into new markets. dSLRs are dying, but imaging is exploding. Nikon and Canon ran into the innovator's dilemma: they didn't want to eat into their core market, so they released crippled products like the Nikon 1 or Canon M-mount. By the time they got into the game, their market was collapsing, they lost a lot of credibility, and Sony had a huge lead. It's probably too late, but we'll see.
Drones, industrial, teleconferencing, and hundreds of other markets are popping up, but Canon and Nikon are conspicuously absent.
Those are exactly quickly going away. Engineers and scientists need to get paid and want job stability, while patents have a 20-year lifespan.
Nikon was on top of the world in 2008 with the D300 and the D90, but the technologies that made those possible were patented years prior. Right now, the curve is with mirrorless where Sony, Panasonic, and (now defunct) Olympus have a 10-year lead on patents. It's also with smartphones, and those technologies bubbling up. In the deep learning/ML battle, Nikon is bringing a knife to a gunfight.
On the other side, they're having massive layoffs right now. When companies have layoffs, usually they let go the bottom end of the workforce, while the top end leave voluntarily.
Nikon is continuing to patent new lens designs, but so is its competition. Nikon has excellent optics, but so does Olympus, Canon, Fuji, or Sony; they aren't exactly ahead of the pack here.
Sigma has really been pushing the edge on optics.