HNHacker News
TopNewBestAskShowJobs

ewjordan

2,337 karma · joined July 7, 2008

submissionscomments
ewjordan··on Project delays: why good software estimates are impossible (2015)
Whatever you're doing, if you're honest with yourself, are you really doing something new and interesting every day/week/month, or are you just digging deeper into what you already know?

I've found that without straight up hopping jobs, it's almost impossible to do truly new things rather than settle into a role where you're too valuable to not exploit for the same old crap you've done before.

I'd still call that doing a good job - businesses need people that can pump out good but similar work. People that always need something new once they master the old thing can be liabilities, depending on the context.

ewjordan··on Barack Obama on A.I., Autonomous Cars, and the Future of Humanity
Sure, and I probably shouldn't have glossed over that. That sort of research is definitely progress, though it's not paradigm shifting in any way. I do think that we are getting past perception slowly but surely, I just don't think we're there yet.

What really doesn't exist is any meaningful stab at unsupervised (or self-supervised) training on completely unstructured inputs or any sort of knowledge condensation/compression, at least for time dependent problems. These are of paramount importance to the way we think, and to what we can do.

There's a lot of trivially low hanging fruit, too - I still have yet to see even a grad school thesis that starts with an N+M node recurrent network and trains an N node subnetwork to match the outputs based on fuzzed ins, and then backs that out into an unsupervised learning rule that's applicable to multiple problems. Or better, a layered network that is recurrent but striated, that tries to push weights towards the lower layers while reproducing the same outputs (hell, even with a FF network this would be an interesting problem to solve if it was unsupervised). These are straightforward problems that would open up new avenues of research if good methods were found, but are mostly unexplored right now.

I could be wrong, if I had real confidence that we were close I'd be working on this stuff, but I'm collecting a paycheck doing web dev instead...

ewjordan··on Be Kind
If there are five equally inexperienced applicants in the wings that will get along with everyone without having an attitude-smackdown first, then that's absolutely a reason not to hire them.

The worst problems I've ever seen on teams spring from people that are arrogant and standoffish, followed closely by people that are too dogmatic about technical minutiae and best practices in the face of situations that require flexibility (my classic example is the prototyping team that spent two weeks setting up an automated testing and deployment infrastructure, even though they only had six weeks to do their game prototype).

I'd take a pleasant junior that's really green but is happy to learn how to code better over a more technically proficient asshole that needs to have manners jammed down their throat any day - tech skills can be taught to a newbie, personality cannot. People have had a couple decades to cement personality habits as junior programmers, so those are mostly fixed, whereas tech skills are super fresh and malleable.

ewjordan··on Barack Obama on A.I., Autonomous Cars, and the Future of Humanity
I'm as bullish on AGI as anyone in the medium term, but deep learning is not even playing the same game as AGI, let alone in the same ballpark or having the potential to achieve it.

Deep learning is still mere perception. It doesn't handle memory or processing, it just transforms input into output, typically trained by Big Data, way bigger than necessary statistically speaking, given the world we live in.

AGI requires super aggressive unsupervised learning in recurrent networks, likely with specialized subsystems for episodic and procedural memory, as well as systems that condense knowledge down to layers of the network that are closer to the inputs. At a minimum. And nobody is really working on any of that yet (or at least succeeding) because it's really damn hard.

That's why everyone in "AI" is rebranding as a deep learning expert, even though deep learning is really just 1980s algos on 2016 hardware - you gotta sex up feed forward backprop or you don't get paid.

Edit: to be fair, robot control is much simpler than AGI, and might be mostly solved with deep learning somewhat soon, I forgot the context of your post.

ewjordan··on Burning Ship Fractal
Structure is not the same thing as information. One is accidental, even if useful, amd the other has to be "designed".

People arguing about the complexity of the human brain (especially those that know a lot about the way it works) fall into this trap constantly, and it drives me nuts: there's a ton of structure, but proportionally negligible information in the brain's programming.

ewjordan··on Qualities of clean code
Over the years I've seen many more unmaintainable atrocities caused by diligent adherence to these principles than I ever have because of ignorance of them.

All the horrid Java libraries that we hated working with 10 years ago were created because of slavish devotion to the single responsibility principle, short methods, DRY, and testability. The absolute worst codebases I've ever seen sprung from an overaggressive concern for extensibility.

I do agree with the guiding principle, that when you're writing code, it should be written for other people to read and work with. But sometimes that actually does mean that a class should represent an object that does more than one thing, and you shouldn't fight it. Sometimes a gritty algorithm really does make the most sense if it's laid out as a 40 line function in one place rather than spread across 5 different classes so each part is independently swappable and testable. Sometimes you really don't need extensibility, you know that up-front, and building it in just pisses off everyone that ever has to touch your code because now they have to go through five or six "Find implementing classes" dances to find the actual code that runs when your interface methods (of which, of course, there's only a single implementation) are called. Don't even get me started on abuses of dependency injection when it's not necessary...

Part of me is glad that these concepts are so commonly discussed, because they really are good things to consider, and can make code much tidier and easier to work with in the best case. But it takes a lot of experience to know when and where you should and shouldn't follow these "rules", and it tends to be much easier to unwind a novice's spaghetti code and tidy it up than it is to pick apart a poorly conceived tangle of abstraction and put it back together in a way that makes sense.

ewjordan··on Some bad Git situations and how I got myself out of them
Fair point. For all my whining, I use both Git and Linux every day, they are incredible pieces of technology, especially compared to what came before.
ewjordan··on Some bad Git situations and how I got myself out of them
People have already mentioned it, but since I made the original comment, I'll pile on and say that Mercurial was (well, still is, but it's out of fashion now) awesome. It matches Git on most axes and blows it out of the water when it comes to usability and "don't shoot yourself in the foot"-ishness.

Brushing aside some relatively minor differences, Mercurial is very similar to Git, except that it was designed by people that don't have an all-consuming disdain for all of humankind.

ewjordan··on Some bad Git situations and how I got myself out of them
Yep, Mercurial's the one. It's a shame that it lost the war, they actually seemed to care about providing a good experience to their users. There are, of course, a few technical differences between it and git, but I personally never felt like they were troublesome enough to make up for the vast difference in usability.
ewjordan··on Some bad Git situations and how I got myself out of them
"Not the friendliest of beasts" is putting it mildly.

Git is basically Linus in a nutshell: abrasive, unforgiving, and behaving like an absolute asshole to any non-expert struggling user.

Sure, engineers should probably get to know its quirks and learn to work around them because it's now ubiquitous in the field, but let's not pretend that there's something virtuous about it. This is a piece of truly terrible software with an even worse UX that just happened to win the PR battle against its superior foes, because...Idunno, Linux, I guess?

ewjordan··on Introducing “Ask a Female Engineer”
That's a bit of a knee-jerk, I think. Men are referred to as "male software engineers", because "men software engineers" sounds absolutely ridiculous.

The asymmetry you're thinking of is when people refer to women as "females" and men as "men". That's a super offensive one, because we usually only use "female" and "male" as nouns when speaking about animals. Better, more specific words exist for humans.

In the current situation, though, "female" is being used as an adjective, which is correct. The only alternative is to (ab)use "woman" as an apposite noun, which sometimes works, but can go terribly wrong as a general rule: "Ask a Man Nurse" vs. "Ask a Male Nurse"

All that said, due to common usage "woman" works as an apposite in many places where "man" does not ("woman president" is fine, "man president" sounds weird).

See http://www.slate.com/blogs/lexicon_valley/2016/02/18/when_yo... for more on all this.

ewjordan··on It’s Tough Being Over 40 in Silicon Valley
See http://www.bls.gov/opub/btn/volume-2/images/2-9-chart2.png

A large number of the people that made up the 3x growth in IT between 1990 and 2000 are just about to hit 40 now. That's a very rough measure, just looking at overall industry size over time (and granted, it's IT rather than startups), but it at least mildly suggests that the demographics are going to start flattening a bit as we hit more of a steady state rather than a period of extreme growth.

ewjordan··on One Hundred Year Study on Artificial Intelligence: 2016 Report
"Ask them what error rate they get on MNIST or ImageNet"

While I agree that Numenta probably doesn't have any sort of full-fledged AI, the human brain does terribly on MNIST and ImageNet compared to the state of the art. So we would fail that test.

Getting stuck on toy problems like ImageNet and overoptimizing solutions that can't possibly be applied more generally (except as dumb preprocessors) is not likely to lead in the most interesting directions, even if it's incredibly useful and profitable in the meantime.

ewjordan··on Motivational Incongruence and Well-Being at the Workplace
I've mostly seen it pushed (as a hard 100% policy, at least) by well meaning but naive people that have done it a few times before, got obsessed with it, and decided it's this week's cure for everything.

The reality is, it will make the team produce fewer bugs. It will also help in knowledge sharing, and in some cases result in tidier designs.

But you'll basically have half the manpower. So unless you actually have serious problems with bugs, knowledge sharing, or code design, it's not going to be worth it. There are other ways to achieve quality, and most senior people can do that very well without someone pairing with them.

ewjordan··on Trial by Jury, a Hallowed American Right, Is Vanishing
The solution to that is pretty simple: vote not guilty, and refuse to budge. If you need any clarification, the prosecution has not done their job. In a criminal trial, you're not there to figure out what happened, you're there to judge whether it has been established beyond a reasonable doubt that the defendant is guilty.

Whether the defense or prosecution made mistakes, if you're left with a reasonable doubt that the defendant is guilty (and I personally consider it the case that inept defense can cause reasonable doubt, unless the prosecution presents an airtight case - a lot of people, including most of my lawyer friends, would disagree with me there), don't vote to convict.

ewjordan··on Habits of highly mathematical people
Oh, they use them. The problem is that each philosopher has a different precise definition that they feel is natural and matches what we "really mean" when we ask deep questions. Since most arguments and conclusions are phrased fuzzily, you end up with the following situation:

Fuzzy premise -> precise logic -> fuzzy conclusion

where each of those arrows leaves so much room for interpretation that you can build literally an entire subfield by arguing back and forth about what the most reasonable mapping from fuzzy to precise and back again might be, even if everyone agrees that the manipulations in between are rigorously correct.

ewjordan··on The gender wage gap in Silicon Valley – based on AngelList data
Your point is spot on: people should always know what they're worth when they negotiate salary at a job, asking for what you can get is the most crucial thing when taking a job.

Both men and women are uncomfortable negotiating up (I've heard that women are much more hesitant about this, but I don't know the research that well and never sat on the hiring side of salary negotiations), and low asks are probably the biggest mistake anyone can make: $10-20k doesn't mean shit to the manager that wants to get you on their team if you negotiate it at hiring time, but if you try to get that bump during annual raise period, that's almost impossible to achieve at any large-ish company (it means either nixing raises for other people on the team or calling in VP-level favors, which most managers don't have the ability to do).

In tech nobody is ever going to tell you to piss off for asking for $10k more than they can offer, they're just going to negotiate you down (if even that - most of the time they'll just say "yes" or split the difference). Now, if you're $50k+ out of line, that could be another matter, but that's why you do some research and don't go crazy with your ask.

The rule for contracting is more brutal, but also a bit simpler: you should always be losing a lot of business because of how high your rate is, somewhere between 25% and 50% is my rule of thumb. If clients are saying "yes" without negotiating or at least complaining, you're definitely charging too little.

Protip: you can look up top salaries at most non-profits online, they have to report them for I think the 15 highest paid employees. Pick a smallish one in your area, look at the engineers there, and you've got an anchor for senior engineering rates. Non profits might not be 100% competitive with for-profit companies, but the better ones have to be somewhat close as far as base salary (they don't offer stock or other perks, usually) if they want good people.

ewjordan··on How to write a 48-hour game in just 2 years (2013)
This is one of the reasons Unity has become so popular in the industry - from a project management point of view, it makes sure that programmers don't start yak-shaving and building tech that's already been built a thousand times by other people.

It certainly has its warts, esp when going into full production, but at least in early development it's really good at making sure you're only building what's unique to your game.

ewjordan··on A look at the 10x paradigm
All of those things being "better" include an implicit "...in the context which these best practices were studied" which usually means large teams of just-OK programmers writing Java code that mostly doesn't need to perform very well but will be changed a lot, and needs to be very consistent in style and pattern so that when an intern comes in and pokes it as his first commit he can't do any serious damage.

You can imagine a million ways those assumptions break down, as well as a million ways they seem to but should actually be stuck to. Knowing the difference is the tough part.

ewjordan··on Why not string theory? Because enough is enough
Humans are neither immune to No Free Lunch, nor able to predict Perlin noise.
ewjordan··on Automatic laundry folding
For the true laundryphobe, you can buy socks in bulk for less than a dollar a pair. Just wear a new pair every day and throw them out when you're done.

I don't actually do this, FWIW, but talking about it is a fantastic way to annoy an environmentalist at a bar. I often threaten to create a startup around the "disposable clothing subscription" idea, 9 out of 10 at the table are horrified by the wastefulness or think it's ridiculous that people consider matching socks to be a nuisance at all, but that last person is so lazy that they'd almost consider it if the price was right.

I suspect it probably wouldn't make a great subscription business, pack/ship costs would make the margins really tight if you wanted to stay around the $1/pair price point (and you'd be sourcing really crummy socks even to get there), so I think you'd have to go pretty far up-market. Who knows, maybe that could work (it looks like a ton of sock-of-the-month premium subscription box things do exist), though probably not for daily disposable socks.

ewjordan··on The truth about deep learning
Yep, there are a lot of innovations, and plenty that people are holding back as well.

None of it involves the word "deep", it's all just bog standard 1980 style nets with tweaks.

ewjordan··on Escaping the Local Minimum: Where AI Has Been and Where It Needs to Go
FWIW, I'm in no way suggesting that evolutionary strategies are the way forward in the search for AGI, though I suppose they might be part of the solution.

My 95% bet is that a random or evolutionary search might be how people optimize an AGI, not how they find it. They'll get there via a minor but innovative twist on the RNN work we've already seen (first people need to realize that backdrop through time is a complete dead end, and they should be looking closer at reinforcement learning methods and figuring out how to bring them online).

ewjordan··on The truth about deep learning
Actually, as a great fan of what has come out of the current deep learning hypefest, I'd question whether "deep" really matters. Most of the great successes have resulted from medium depth nets using the same shitty backprop algorithms that have been known for decades.

"Deep learning" is a red herring. They're just doing exactly what the shallow learning pioneers told them to do twenty years ago, just with more computing power.

ewjordan··on Escaping the Local Minimum: Where AI Has Been and Where It Needs to Go
I nodded along with pretty much everything you said, so I think we're arguing different points but not necessarily disagreeing.

I agree, evolution's scale is colossal, but restrict your view to the tiny bit of the evolutionary machine that is implicated in figuring out how to arrange neurons (which evolution "discovered" previously, to be fair, with incredible effort) in a fashion that renders their behavior intelligent, and it becomes a vastly smaller effort.

These beings are all rather large, at least worm-sized or greater. Their lifecycles are measured in days to years, not minutes or hours. And they only had 250 million years or so to make the leap from "bunch of neurons hard-coded to do specific jobs" to "generally intelligent arrangement of neurons capable of higher level thought". All of this cuts the scale by at least 5-10 orders of magnitude compared to what you correctly point out as the overall colossal scale of evolutionary computation. And to me, it says that compared to all the other amazingly difficult stuff that evolution has discovered, intelligence was a damn easy find, especially since there's almost no fitness benefit - hell, worms probably do better than we do as far as evolutionary fitness goes, they're tiny, numerous, and reproduce like crazy.

My money is on the fact that human intelligence guiding a search cuts another couple orders of magnitude off of how much of a long-shot discovering an intelligent algorithm would be via a random search. I could definitely be wrong, it's very hard to be precise when you're vaguely arguing about orders of magnitude, but to me it doesn't seem like the algorithmic "magic" in the brain can possibly be very complex if evolution was able to get there, there just aren't many other circumstances where evolution stumbles on something that damn clever.

ewjordan··on Escaping the Local Minimum: Where AI Has Been and Where It Needs to Go
No, we don't even need to be very clever: evolution is no better able to evade weak local minima than any but the dumbest search strategies humans use, and it quite easily discovered a neural algorithm that lead to intelligence when scaled sufficiently.
ewjordan··on Do men or women have more brothers?
This is a late reply because I missed this until now, but if you accept the typical parameters of a statistical thought experiment, you should assume that the permutations are equally likely. The real world nastiness of biology and society fouls that up, but thought experiments like this are never intended to be about not picky real world details, boy/girl is always assumed to be a coin flip. (Not to say that real world results are not interesting as well)

And there are a half dozen arguments in this thread that claim to disprove this result even if they accept the coinflip hypothesis - it's really those that I take issue with.

ewjordan··on Why David Hume Is Popular Now
"...the existence of the self, a subject that’s still scientifically unsettled."

Shit like this, philosophers, is why scientists don't (and can't, and shouldn't) engage. The words to use are "scientifically undefined", not "scientifically unsettled" - you can't just argue over definitions for centuries and then blame science for that fact that none of you can agree on what you're talking about, if anything.

You enter the realm of science when you have a testable hypothesis, not just some vague idea that reeks of dualism that you can't conceive of not being a "real thing", like consciousness or self.

ewjordan··on Do men or women have more brothers?
Like tempestn noted in another comment, enumeration is straightforward for small families, and makes it crystal clear that for any family size N, boys and girls have the same average number of brothers:

  Gender:     B  G
  # brothers: 0  0
  B avg # brothers: 0
  G avg # brothers: 0

  Gender:     B,B  B,G  G,B  G,G
  # brothers: 1,1  0,1  1,0  0,0
  B avg # brothers:     (2 + 0 + 0) / (2 + 2) = .5
  G avg # brothers: (0 + 2 + 0) / (2 + 2) = .5

  Gender:     B,B,B  B,B,G  B,G,B  G,B,B  B,G,G  G,B,G  G,G,B  G,G,G
  # brothers: 2,2,2  1,1,2  1,2,1  2,1,1  0,1,1  1,0,1  1,1,0  0,0,0
  B avg # brothers:     (6 + 6 + 0 + 0) / (3 + 6 + 3) = 1
  G avg # brothers: (0 + 6 + 6 + 0) / (3 + 6 + 3) = 1

  Gender:     B,B,B,B  B,B,B,G (x4)  B,B,G,G (x6)  B,G,G,G (x4) G,G,G,G (I collapsed the different permutations here)
  # brothers: 3,3,3,3  2,2,2,3       1,1,2,2       0,1,1,1      0,0,0,0
  B avg # brothers:     (12 + 24 + 12 + 0 + 0) / (4 + 12 + 12 + 4) = 48/32
  G avg # brothers: (0 + 12 + 24 + 12 + 0) / (4 + 12 + 12 + 4) = 48/32
In a family of N children, both boys and girls will have, on average, (N-1)/2 brothers.

Arguments for unequal results are slipping in assumptions that make the observed distribution different from the enumerated one.

It's an exercise for the reader to prove that this equality holds for higher N, it's quite easy (the way I formatted the arithmetic shows how to match up terms).

ewjordan··on Do men or women have more brothers?
Come on now - thought experiments about the sex of children always ignore the real world complications that make them different from a coin flip (stopping conditions, genetic predisposition, environmental effects, etc.). And after making those standard thought experiment assumptions, the problem is 100% napkin-worthy.
← PreviousPage 6 of 18Next →