johnhess

Granted, there have long been wysiwyg editors and this is a relatively simple part of the job of a web dev, but it's just another example of software eating the world.

The original courses that edX hosted were authored in a rather tricky XML format. Authoring and QA in that system was time consuming for developers and borderline impossible for non-technical users.

The article pays lip service to the fact "There's no way to A/B test hiring decisions" but then goes on to describe a handful of rules to "build an effective organization, regardless of size or resources." The thing is, there's no evidence that these are effective and definitely none to suggest they're generalizable to organizations of every size. There's an appeal to the author's authority as an Amazon VP (admittedly a hell of a credential), but that's not evidence of effectiveness.

The fact is, hiring is tough to learn. It's made especially difficult because your feedback comes at least 6 months after you make the decision. I've had the good fortune to be in an organization which grew from ~10 employees this time last year to ~80 now, so I've gotten to see how a few dozen hires panned out in a short period of time. I learned a lot through that. Even so, as a professional data geek, I couldn't say with a straight face that I "know" anything about hiring. I do my best day in and day out, but saying you've got hiring all figured out is the professional equivalent of knowing "What (wo)men want." It's a sign of confidence, but not necessarily knowledge.

Don't get me wrong, in the absence of quality data, I'll take the wise words of someone who has been there and done that. But I'd much rather some empirical proof or at least some "YMMV" humility from those who purport to know.

Many good instructors can't draw a line between KA (or any other disruptive solution) and the absolute best, gold standard, scientifically-proven best way to teach (usually their favorite methodology).

The fact is that he's better than most of the teachers out there. It's a rising tide: if your teacher has a Presidential Award for Excellence in Mathematics and Science Teaching you probably don't need Sal Khan.

1. Attacker prompts me (or my grandmother) for login name. 2. Attacker gives login name to bank. 3. Bank serves proper image to attacker. Attacker stores image. 4. Profit.

"it becomes possible to measure the properties of individual raindrops greater than 0.5mm in diameter due to the low concentration of such drops in naturally occurring cloud systems and the overwhelming dominance such drops have on the measured radar reflectivity when present in a field comprised of smaller particles."

It's easy to distinguish an airplane from something with a similar sized signature, say an insect, since the airplane moves at 1000knots instead of 1, but you have to be able to see it through the noise. That's tricky. If you use a radar like this (high power, really narrow beam) you might be able to see a stealth plane. But, you'd have to know where to point it. Therein lies the challenge.

Clearly, the metric is far from perfect -- you can tell that without his analysis. Year-over-year variation is high, indicating that there's likely a lot of noise in the measure. That said, year-to-year scores do correlate. This means that the measure is at least a little bit reliable in the sense that if you measure the same thing twice you'll get an answer that's close to the original.

All he's shown is that these are "noisy". There's no reason not to use a noisy metric, you just need to know that's what you've got, and make responsible decisions.

For example, don't fire teachers who rank poorly, but do review their performance. In that case, we'd find that the "Worst Teacher" is underperforming on this metric because it wasn't designed to test great students, and we can move on. Another example of the same phenomena in education is looking for teachers who help students cheat exams. When there are suspicious correlations found, administrators investigate.

One last thing: What he doesn't investigate -- and this is much more important -- is the (lack of a) correlation between value added ratings and other metrics (say, peer evaluations).

hit people upside the face with it

YMMV.

Also, vonsydov had a good idea. I know a lot of MIT programs let you take a 1 year deferment. Maybe you can take a cut at the startup, and if you go belly up, go to school. If you end up a billionaire, buy vonsydov a beer.

Yes. Yes. Yes. One thousand times, yes. If you're asking me to dedicate my waking hours to something, be clear about what that something is. Hint: It's not your corporation.

I've interviewed at several companies, and the only ones I rejected out of hand are the ones that are just "doing business" and have ceased to chase some vision.

Some companies do really cool stuff, but have no vision (e.g. most of Lockheed Martin) whereas there are companies who do something seemingly less exciting (e.g. Yelp) that have a vision for how they're going to change the world.

In Yelp's case, they're helping great businesses thrive and be found and holding bad companies accountable. As an Aero guy, I'd be predisposed to Lockheed, but frankly, when I talked to Yelp, that kind of thinking made all the difference.

If you ignore the dip, it looks like it might be nothing at all or just a "backlog" of would-be searchers from Christmas.

This is on the front page now, but holy crap is it beautiful. This is everything data analysis should be.

When you're trying to do meaningful data analysis, you really have to understand your dataset. Fancy math can't substitute for domain expertise. Think long and hard about what's in the set, what the causal connections might be, and how an "expert" in the field might approach the problem.

The guys over at OKCupid are awesome at this. Check out this post to see what I'm talking about.

http://blog.okcupid.com/index.php/dont-be-ugly-by-accident/

Their advice on taking a good picture is just about exactly what my professional photographer mother recommends. Good stuff. But, the way they analyzed and presented the data shows (a) exactly how powerful putting numbers on something subjective can be and (b) that they know their domain.

If you read through the other posts (do that), you'll see that they have a solid understanding of their dataset. They know what to look for, namely photo attractiveness. They know how to get good data on that (the dependent variable) and they know which independent variables probably matter the most.

Throwing math at a complex dataset can be useful (e.g. bayes spam classifiers), but if you really want to do something that will work well or "speak" to a client, invest a bit of time in understanding the field.