greatzebu

Wouldn't that make them entirely beholden to search engines that rely on ad revenue to pay Chrome? It doesn't seem to change the underlying incentives very much.

Grades are also fraught with bias, and in a lot of cases measure compliance more than mastery of the material.

The tip is for the person coming up with the rule--even rules that seem very simple will be harder to figure out than you expect.

It never attracted a giant user base. I miss Reader (especially the pre-G+ integration version) and I wish it was still around, but it's not mysterious why it got killed. If it had hundreds of millions of users it wouldn't have died.

I think this should be read as "kids under 6 failed [to perform better than chance] every time"

The principled way to avoid bad worst-case performance for quicksort is to select the pivot using a median of medians approach. That gives you good asymptotic performance even for adversarial inputs, but in practice it tends to be slower. Which fits well with the theme of this article.

> A "Pareto optimal" solution is one which, given all the possibilities of action, produces the best outcome for both parties (with some negotiable surplus).

This isn't quite right. In a Pareto optimal solution, there's no way to improve things for one party without making them worse for another party, but there's no guarantee that the outcome is good for everyone. "I get everything, you get nothing" is a Pareto-optimal way of diving things up.

The network looks relatively anemic at first glance. I would guess that this will be a LINPACK monster, but it'll be difficult to scale applications that aren't pure linear algebra up to the point that justifies a single machine this size.

"Lede" is supposedly a holdover from the time when news was printed using actual metal lead type, and "lead" was accordingly shorthand for type itself. The alternate spelling avoids a potentially confusing homograph.

I'd say that the two communities here are just people working at opposite ends of a continuum of applications that run on large clusters. The HPC community is all the way at the end with tightly coupled applications, low data to computation ratios, and diverse communication patterns. The big data community is characterized by giant data to computation ratios, highly constrained and regular communication patterns, and loose coupling.

The fundamental problems are similar (fault tolerance, load balancing, scheduling) but the best approaches depend on where you are on that continuum.

One possible explanation: the cost of initial IO to write your data is amortized over that data's lifetime. The early-delete penalty ensures that Amazon always makes enough on your data to justify the cost of writing it into their system in the first place.