aothman

TS does not fight the math of venture capital the same way Indie.VC did. My understanding from Einar (and the reason I made an LP commitment) is that the same extreme power law of returns that characterizes traditional early-stage venture is at play in Micro-SaaS: the exits are an order of magnitude smaller but you also buy in at prices that are an order of magnitude smaller.

It may be telling that, in response to my "Seed investments follow an alpha < 2 power law" paper, Bryce (whom I have never met) posted something dismissive on twitter whereas Einar reached out to me to discuss how he could validate a similar hypothesis for his own investing.

Of course there is a way to broadly index among all non-negative seed investments: by broadly indexing among all seed investments.

The fraction of money-losing investments in a population will affect, for instance, whether we would expect the typical investor making five investments at random to make or lose money. But regardless of whether losers are 10%, 50%, or 90% of the investment pool, if the winners are drawing from an unbounded mean power law then broadly indexing raises an investor's expected return.

Your last paragraph is a misinterpretation of this paper. You are interpreting the result as saying "Investors would benefit from broadly indexing at seed". The paper instead says "Seed investors would benefit from broadly indexing".

Study author here.

You can think of the AngelList investment data as being split into three roughly equal-sized groups: markdowns, markups, and no valuation updates.

The reported IRRs are actually relatively high and the return multiples (which are compounded IRRs) are relatively low. That's because there are lots of one- and two-year-old companies in the dataset and---as we show---IRRs and investment durations are negatively correlated.

HiScore author here. HiScore allows domain experts to easily create and maintain scores. It is currently being used by a major environmental non-profit and by IES, a startup that assesses the safety and sustainability of fracking wells.

The algorithm itself is a mashup of a paper I wrote for AAAI: http://www.cs.cmu.edu/~aothman/splines.pdf and this paper: http://link.springer.com/article/10.1007/s10543-005-0028-x

Seems strange to write an article complaining about the Harvard bubble when the only reason it's getting printed in the Paris fucking Review is that it's about Harvard in the first place. People have weird/terrible/difficult college experiences all the time, but the guy who went to Iowa State never gets a similar forum to talk about himself.

A lot of people at Harvard were uncomfortable there. I wasn't. I loved the place and made great friends. It was a safe place where I could challenge myself intellectually. And the plurality of Harvard students aren't wealthy legacies, they're striver upper middle class kids that worked their asses off in high school.

Uncle Tom's Cabin is an awful book. First off, it's boring and damn near unreadable (it was one of the only assigned books I never made it through in college). But in a larger sense, the slaves are "heroic" and "emotionally nuanced" only in the sense that HBS makes them fulfill a racial type: sympathetic, penitent, long-suffering Christians. They're treated more as people than as property, but more as caricatures than as people.

The interesting contradiction of UTC, to me, is that it had this enormous significance to history despite being terribly written. As a modern reader, I couldn't get any emotion about the book other than it being terrible. James Baldwin trashes the book brutally but fairly in his great essay "Everybody's Protest Novel": http://www.uhu.es/antonia.dominguez/semnorteamericana/protes...

My favorite Friedman-ism is this one: "I had lunch with a group of professors at the Hong Kong University of Science and Technology, or HKUST, who told me that this year they will be offering some 50 full scholarships for graduate students in science and technology. Major U.S. universities are sharply cutting back."

http://www.nytimes.com/2010/01/13/opinion/13friedman.html

50 graduate scholarships is roughly the equivalent of a smaller department at a US research university.

As an AI researcher, I think obstacles like "not having your robot fall over all the damn time" are a little more immediate than robots having a nuanced understanding of ethics. I can understand why this stuff is fun to think about and debate, but it's just not relevant at all to where AI is going to be for the next 50 (or 100, or probably 200) years.

"Proverbial Stanford" coincides a great deal with "Proverbial Harvard" - both are wealthy private schools that admit the very best students and have society's bias towards the well-to-do sons of well-to-do fathers. If you're looking for a school to contrast with Harvard, Stanford is a poor choice.

Furthermore, actual Stanford isn't doing any damage to actual Harvard. The data I've found suggest that 70+ or 80+% of undergrads admitted to both Harvard and Stanford pick Harvard.

Sources: http://college.mychances.net/college/tools/college-cross-adm...

http://mathacle.blogspot.com/2008/06/harvard-yale-princeton-...

As an elitist alum, I don't like these programs one bit. I can't help but feel that they are, in a small but meaningful way, watering down the value of my degree. And even though it's petty, I'm chagrined that my diploma features English rather than Latin text.

Harvard's Extension School was designed to teach the greater Boston community, and I think it should be a vehicle to improve town-gown relations, by convincing locals to not perceive Harvard as "the other". I certainly don't think it should have as part of its mission handing out Masters degrees to people from Kansas over the Internet.

If you're interested in this kind of stuff, Ken Pomeroy (kenpom.com) runs a fantastic basketball analytics site. He's a big proponent of what are called "tempo-free stats", which aim to filter out issues of playing speed from scoring (a team that plays quickly will score a lot of points, nearly independently of whether or not they are winning). Tempo-free stats instead count possessions; one interesting statistic is that the average team this year in college basketball produced 1.01 points per possession - such a tidy figure to emerge from the chaos.

In terms of predictions, one of the most interesting teams this year is Kansas (http://kenpom.com/team.php?team=Kansas). They've only lost twice but have a large number of narrow home wins. Depending on how your algorithm treats those wins they either look like a team that will struggle to reach the sweet 16 or like a potential national champion.

Pittsburgh's 1985 win (in the middle of the steel industry collapse) prompted a lengthy screed about how cumulative ratings are assembled from a UW psychology prof: http://www.briem.com/files/Loftus1985.pdf

Pittsburgh keeps winning most livable city awards today for the same reasons it won in 1985: it's a city that rates good or very good in everything and bad in almost nothing.

As an AI grad student, this kind of sensationalism is somewhere between a minor irritation and a serious threat. AI always has had a severe problem with over-promising and under-delivering, and I'm of the humble opinion that until you're actually shipping the most awesome thing in the world you should keep your mouth shut. If the first thing people associate "AI research" with is "disappointment", that hurts everybody (particularly, NSF funding).

"Brain-based" AI should stay in the dark ages. Optimization-based AI is the present and the future.

(That said, if you want to talk about your sweet computer vision system that's "coming soon", go right ahead. Just don't call it AI.)

"Your time to break even is quite a while on a per-article basis" - and that's the crux of the matter entirely.

As per this WSJ article http://blogs.wsj.com/digits/2010/08/12/where-did-demand-medi...

if demand media is allowed to amortize out their costs over years (under the assumption every piece of "How to bathe your gerbil" will keep bringing in revenue for the next half-decade) they become strikingly profitable, especially for a company that writes words on the Internet. If not, they're hemorrhaging cash.

Given Google's recently renewed focus on providing quality search results, I'd take the under on all of these companies, mahalo included.

http://catandgirl.com/?p=1163

"Every once in a while a window appears - and you realize the past is just the present with different technology and funny clothes"

My research involves making electronic agents to mediate interactions in markets, with one application being to prediction markets. One of the things I've discovered is that prediction markets where prices influence events are significantly more complicated from a theoretical perspective than a standard market. In such markets it can be difficult to predict how rational agents will act, even if they care only about their risk-neutral monetary payout and not about the events in question.

For instance, one of the new trends in prediction markets has been corporate elicitation (e.g., Inkling Markets YC W06). For instance, a market might ask a pool of employees "When will product X ship?". Now consider the employee's problem. He knows the product will be delayed until November at the current pace, but that if managers see high November probabilities they'll cut back on the project and re-focus development efforts (or, perhaps instead, shift some staffers over to get the project done faster). How should the employee trade in this market?

If your answer is "it's unclear" then you're exactly right. The self-referentiality of prices (they cause actions which then define prices) makes the problem much harder. It's actually possible to design markets which are so self-referential that any trade will be the right one (the market prices are self-predicting prophecies) or that any trade will be the wrong one (e.g., if there's a high price on terrorist attack Z then the government will always spend enough money to prevent Z from happening).

In fact, the only way around the paradox is to have the prices in a prediction market not matter at all - for the managers to take the same course of action regardless of the prices they see in the market. But then why run a market in the first place?

The whole thing is an interesting and seemingly fundamental perversion of the way we normally think about prices. And again, this is without any malicious intent on the part of participants, which will certainly only further confuse things.

Chuck Klosterman wrote a wonderful essay (it's in his collection "Eating the Dinosaur") about laugh tracks, and more broadly, about the way we use laughter. His thesis was that, especially in the mundane interactions among strangers that populate daily (US) life, laughter is a nearly continuous stream that reassures everyone that both they and everyone else around them know what's going on.

Next time you're in a conversation with a group of people you don't know, stop and listen for the laughter - it will be hard to hear because you're so used to tuning it out, but it will be shocking in its frequency when you finally pick up on it. (NB: pointing the laughter out is a good way to alienate yourself from strangers.)

Klosterman talks a bit about his essay here: http://blogs.wsj.com/speakeasy/2009/10/17/chuck-klosterman-o...

Looks impressive, particularly if those are real trading results.

One concern I immediately have is overfitting, particularly for claims about how various difficult values have been optimized to be the "best possible". It looks like the parameter space in use is truly enormous and so it would be very easy to come up with hypotheses that perform fantastically on your dataset but terribly in real life. This seems like it would be a first-order concern, while the ability to run tests in a single day seems second-order if those tests are producing garbage outputs.

I think ultimately the profit cut they take - a practically usurious 50% - is too large to be an equilibrium price for what they do. There is a long-term sustainable biz model in what they're doing, but it involves taking a much smaller cut, just because they're not adding enough value to justify a huge cut (they're not exactly investment banking). I expect that in the near future market competition (particularly Google deals) will drive the middle-man profits down to something pedestrian.

Cool demo, but from a practical perspective there's absolutely no reason to use GAs to solve a non-linear optimization problem. I know people like them because they have a really pleasing and intuitive backstory, but as a grad student in AI I can tell you that they suck at actually solving anything. Mostly this is due to the fact that GAs take what is already a difficult, non-linear problem (the problem you're trying to solve) and immediately, explosively, complicate it (what's your mutation rate? what are the chromosomes? how are crossovers handled? how are you deciding the answers to these questions?)

Touretzky is also one of scientology's biggest online critics:

http://www.cs.cmu.edu/~dst/Secrets/

I believe CMU's lawyers are well-acquainted with responding to legal threats against him.

The point about turning out a generation of clones is spot on, and ultimately the cruelest irony of the whole thing. The best way to get into an elite college is by standing out as an individual; the colleges asian parents desperately want their kids to attend deal with the "asian clone" thing by rejecting the lot of them. The asian kid with a 1560 SAT and state violin awards (probably) isn't getting into Harvard, but if he had substituted kicking field goals for every minute he practiced violin...

Why Austin (and not New York or Boston)? What was it like doing YC and then not staying in SV? Was there any disagreement among the founders or investors about moving away from the valley?

For solving local search problems, I use tabu search (hill climb with a "recently visited" list) or beam search (simultaneous hill search). Both are simple techniques that show remarkable emergent behavior. Many search problems are better phrased in terms of numerical optimization or what have you - if your problems maps a continuous space to a continuous space there's probably a standard numerical technique that solves it better than a local search hack.

For Machine Learning type applications, SVMs are very popular. Briefly, both sufficiently deep neural nets and sufficiently dimensional SVMs are arbitrarily expressive, but SVMs give you a better perspective on what is actually happening with your problem. If you're interested in Machine Learning, you should check out Andrew Moore's very well-written tutorials: http://www.autonlab.org/tutorials/list.html

GAs are just a really complex version of a local search algorithm. The problem with them is that they're just too complicated - you're trying to solve some non-linear problem, and your first step is to introduce several more non-linear problems that also need to be solved (picking chromosomes, mixing, population size, etc.)?

NB: This article (and several more like it) was in January's wired.

As an AI researcher, I get suspicious when I see anyone talking about Genetic Algorithms and Neural Nets. These are techniques that current researchers simply do not use (Neural Nets are used very sparingly, GAs should never be used at all). They make up for their technical failings by being approachable, particularly for journalists. In short, these methods intuitively sound like they should work much better than they actually do.

Sounds like he was in the right place at the right time. I think it would be a mistake to read anything into his story other than "be really lucky".

EDIT: There were lots of people just like him that weren't crazy-successful. It's wonderful that he put himself into a place where he could succeed, but that's only necessary, not sufficient, to realize that success. He deserves credit for buying the ticket and taking the ride, but beyond that it's luck.

A lot of this has to do with the way we "score" forecasts. Applying some kind of uniform weighting over forecasts is the natural, and wrong, way to think about things.

Nassim Nicholas Taleb (black swan, fooled by randomness) has written about his trading strategy. At his fund, he consistently takes positions that predict extreme events, and he's wrong almost all the time, consistently producing grinding, negative returns. He's only been right a couple times, but when he's right he's really right, making enough money that he doesn't need to make money anymore.

http://www.youtube.com/watch?v=I6IQ_FOCE6I

Just a couple years too early...