svdad

Love the storm trooper.

Specifically in the Gates building, I believe... but what about the tribe across the road in Sequoia Hall?

I couldn't agree more with this. The ML courses that Ng and Koller teach are really missing a lot of the statistical tools you need to do real-world data mining and ML.

My experience: I had basically zero math background, but I took ML with Ng and probabilistic graphical models with Koller, and later was a TA for Ng's ML class, during my Masters' degree and thought I was all set to go into machine learning jobs. To my surprise, I consistently found myself in interviews stumped by questions from basic stats, particularly significance testing, which people with more traditional stats backgrounds assume is basic knowledge (and it should be), but which wasn't taught in any of my ML classes.

I'm in a job now that involves some machine learning, but the ML component is 50% marshalling data (formatting, cleaning, moving), 40% trying to figure out how to get enough validated training examples, and 10% thinking about the right classifier to use (which someone else already implemented). Which to be honest is not very interesting.

So yeah, becoming a real data scientist is hard, requires a lot more knowledge than you get in one ML course, even from Andrew Ng, and the reality of the work often doesn't make it some dream career. And the competition for jobs isn't from other people who also just took that course -- it's from PhD statisticians and statistical physicists who might have taken one ML class to show them how to use all the mathematical tools they already have to do the new hot thing called machine learning.

> However, a defining quality of successful projects I've been involved in is that there is one person driving the overall design and that person runs things closer to a dictatorship than a democracy.

Yeah, that's been my experience too, but I keep running into people who think everything runs better if the whole team decides and everyone gets a say. OK, there's definitely merit in getting input from what one hopes is a bunch of smart people, and it also helps morale for everyone to feel like they're being heard, but at the end of the day I think there really needs to be one or at most a few people who make the final decisions.

Oddly, I just finished reading that. Can't say it's terribly prescient -- it was published just three years ago -- but it does echo lots of things that are just happening now.

> You get the simplest code in the least effort if you solve each problem at the appropriate layer. People who never look down into the layers below them end up with lots of unnecessary duplication and cruft.

Well said. True in general, but especially when you're working on some middle layer of a large scale project, it really pays off to understand what's going on at least on the adjacent layers.

Yeah, I think I misspoke somewhat, but not entirely. The Amazon comp structure, at least for the first four years after you join, as far as I know is something like this (and take this with a grain of salt, I haven't worked there):

Y1: Base + large annual cash (non-perf) bonus + 5% stock vest + optional perf stock bonus

Y2: Base + large annual cash (non-perf) bonus + 15% stock vest + optional perf stock bonus

Y3: Base + 40% stock vest + optional perf stock bonus

Y4: Base + 40% stock vest + optional perf stock bonus

So yeah, they don't adjust the base upward, but they do give you extra cash in the first two years to compensate for their back-loaded stock vesting schedule. That's what I meant.

In SF Bay area, my experience in the last year or so has been that salary comp is pretty much the same between start-ups and big companies, the only exception being super-early-stage startups (i.e. pre-series A) where they really don't have the money yet. The difference is that at bigco, they add bonus and equity grant that have real value, while at startup, they add some % of equity with unknown value. Companies like Amazon and Netflix are also different because they adjust the base salary up to compensate for their different equity award policies.

This is awesome. I'm going to lace all my cables now.

In these kinds of applications, what do they use to attach the cables to something? It looks like on the rover they have special tie points built onto the rover surface -- is that what is done in other applications too?

I think this is fairly trivially explained in most cases by the fact that Wikipedia is an encyclopedia. Almost by construction, nearly every article starts with "X is a Y" (modulo grammatical niceties) where Y is a more general class to which X belongs. In cases when the article doesn't follow that form, the first linked term is usually still explanatory or definitional in some sense and hence more general, and cases where that doesn't apply (e.g. Obstacle) are unusual enough that they don't generally break the cycles. There are no "Wikipedia axioms" that I know of, so this is pretty much guaranteed to end in a cycle looping through a limited set of terms used to talk about abstract concepts. Hence Philosophy.

Seconded -- OneNote was/is an amazing piece of software. I took all my notes for the CFA on it in 2008, and I used it to great effect apartment hunting off craigslist that year as well. My tablet died a couple of years ago and I haven't replaced it, or I'd probably still be using OneNote.

I've found this as well. Just blindly writing test cases doesn't work so well unless you've already understood the higher level operation of what you're trying to build, and obviously does tend to slow down development.

Very. It's just how my mind puts things together, not something I can try to do or not do. I tend to see programs in blocks and patterns interacting rather than sequences of logical operations, probably why I tend to prefer the object-oriented paradigm over others. I get bogged down when I can't visualize the interactions. It isn't just programming, either -- working with derivative products in finance I formed very detailed visual mental models of how they operated.

In fact, most people outside finance would be surprised at how much sophisticated analysis and interactivity you can build in Excel. Once you start digging under the hood it's a powerful platform. The only addition you need is a pricing feed e.g. Bloomberg.

Quite right. Best line from this article: "If you can’t stand to see your child unhappy, you are in the wrong business."

What I wonder, following this story this week, is how the software quality controls at a place like Knight compare with those for life-critical systems like those in, e.g., aviation.

On one hand, you'd think the QA in finance would be pretty solid, considering that the survival of the company could be at stake (witness Knight). On the other hand, I have a feeling that even there, people just don't take it that seriously.

Would love to hear from anyone with more experience writing software for these industries.

Emphasis on the "aspirationally" -- it really appears not to be how most companies do things, despite overwhelming evidence that doing these things work.

Yeah, I think this is right. And NBC actually did this for the soccer World Cup, IIRC -- you could buy a package subscription from NBC letting you watch all the games online.

I have to imagine there's a market for pay-as-you-go subscription to things like Olympic events, same model as buying movies on Amazon or Google Play. I know I would pay.

As it is, try getfoxyproxy.org.

No, "preference" implies intention and active choice. A preference does not exist in the absence of a human actor to have it.

That there is a majority is a statistical fact. Whether anyone has a preference for male employees is the controversial question you refer to.

Thank you for such a detailed response. It's particularly timely as I'm trying to figure out how to introduce my kids to mathematics.

Granted that there is a lot wrong with the current system, I'm curious what you think about the specific criticisms levelled against Khan in the OP's article. Is there any merit in the charge that he introduces the equality operator without enough explanation? Or that there ought to be more discussion in his lessons about the meaning of the decimal point?

Thank you. I had no idea what the "secret" was myself.

What would be meant, in the original article, by the "multiplication algorithm"?

I know that's a bizarre question, but actually I don't think I could explain what that means to someone myself, so I'd better ask.

God I wish that attitude were more common.

The incentives also don't line up on the buy side. As a buy-side fixed income investor, you can buy a 10-yr US Treasury (AAA) yielding X%, or you can buy a slice of the 10-yr AAA tranche of the Abacus CDO yielding X+0.50% (made-up but representative numbers). You're suspicious of the long-term performance of Abacus (it's 2007 and you're already hearing rumblings about problems in the housing market) but your bonus is based on the quarterly performance of your portfolio. It's pretty clear what most institutional investors did in this scenario.

Full disclosure: I worked on a Magnetar trade; not the one that was in the press, but another deal basically identical. And I bought, sold and analyzed Abacus.

All of these transactions are explicitly caveat emptor. My grandma couldn't go out and buy a piece of ABACUS 2007-ACA; sales are restricted to "qualified investors".

Of course we all know that's basically meaningless, at least the assumption that "qualified investors" always understand what they're doing. I know from experience that "qualified investors" in this kind of deal mostly don't do any serious due diligence. [On Abacus, it was practically impossible anyway. There were O(1,000,000) loans underlying the deal and there is no way you'd have been able to get the loan details for even a significant fraction.]

But I wonder whether we're expecting more from GS/MS/etc. than we expect from our local car dealer, and whether the remedy should be any different. That is, everyone is told to be suspicious of used car dealers, so you poke around, kick the tires, look for leaks, get a mechanic to look it over, read the repair records, etc. before you buy. If it still turns out to be a lemon, and you discover the dealer told you something explicitly false about the car, you have a legal remedy; otherwise (IANAL) I don't think you do. But you have a social/market remedy, i.e. don't buy from that dealer again and tell your friends not to buy from that dealer again. Likewise, if you find a dealer who always tells it straight and prices it fair, you go back to that dealer the next time and tell your friends about him.

Why isn't that the right remedy for these CDO deals as well? Sure, if GS said something provably false about the deal, there's a fraud case. But if (as I suspect is more accurate) they just sold it for what it was (namely an extraordinarly complex and impossible-to-price derivative financial instrument, offering an attractively high risk-weighted rate of return), and the buyers didn't bother (or have the computational resources, as almost none of them did) to do their own pricing analysis, is that the fault of GS? If you buy a CDO and lose your shirt, well, don't buy CDOs again and maybe don't buy from GS if you didn't like how they sold it.

Only briefly. :)

My wife and I went out for a celebratory dinner recently when our net worth finally turned positive after years of grad school.

couldn't agree more. "i only made a little money" -- enough to have started angel investing in 2004 after exiting paypal.

enough money to live in one of the most expensive places on earth, probably without having to work too much? check. a family and (apparently) happy marital relationship? check. moderately successful times working at important companies? check.

Sure, it's good to be ambitious and set high goals. But the dude needs to get a grip. If he wants to achieve more, fine. But there's no universal standard by which he's a "failure". At the end of the day it's up to him and he'd do well to admit it.

As an old (37) guy with experience in a couple of start-ups, there's another benefit to having older, more experienced people on the team -- if you get the right people, they will have seen the problems and done the work before, and will save you a lot of big mistakes up front.

The start-up I currently work for was founded by a bunch of very smart, very young engineers over four years ago, and we're still dealing with all the bad engineering decisions they made -- decisions that look obvious to an experienced engineer but aren't obvious to someone without the requisite cuts and bruises. Having even one seasoned engineer on the team then would have saved the company a lot of pain over the subsequent years and would have put us in a much better position now.

Isn't that Morningside Heights? The southern boundary depends on who you talk to... can go down to 110th or so. I grew up on 119th and Riverside. LOVE that area, even thirty-odd years ago. Harlem doesn't start until 125th on the west side.

Anyway, the west side of Harlem, the other side of Morningside Park, has gentrified significantly in the last 20 years and is also by all accounts a pretty nice place to live nowadays.

Washington Heights is also "up-and-coming", up north of there.

"It's normal that things are hard."

This can't be said enough. Everything sucks in some way -- and I mean everything:

* working for a big company (too much politics, can't get anything done)

* working for a start-up (not enough people, not enough money, too much pressure to get product out the door)

* being single (feeling lonely, no sex)

* being married (having to deal with someone else's issues all the time, not enough freedom to do your own thing when you want)

* living in Tahiti (food is too damn expensive because it has to be imported)

* living in San Francisco (terrible weather, house prices, too many hipsters)

* Jira

* Trac

* Vim

* Emacs

* etc.