Profile cover photo
Profile photo
Michael Gebetsroither
Michael's posts

Post has shared content
Now Anyone Can Deploy Google’s Troll-Fighting AI

On Thursday, Jigsaw and its partners on Google’s Counter Abuse Technology Team released a new piece of code called Perspective, an API that gives any developer access to the anti-harassment tools that Jigsaw has worked on for over a year. Part of the team’s broader Conversation AI initiative, Perspective uses machine learning to automatically detect insults, harassment, and abusive speech online. Enter a sentence into its interface, and Jigsaw says its AI can immediately spit out an assessment of the phrase’s “toxicity” more accurately than any keyword blacklist, and faster than any human moderator. The Perspective release brings Conversation AI a step closer to its goal of helping to foster troll-free discussion online, and filtering out the abusive comments that silence vulnerable voices—or, as the project’s critics have less generously put it, to sanitize public discussions based on algorithmic decisions.

Post has shared content
Is Crime Genetic? Scientists Don’t Know Because They’re Afraid to Ask

"...heritability. A term that is much maligned in disciplines like criminology and often serves as a wellspring of confusion. Humans differ in height, weight, personality style, and behavioral tendencies — not everyone is nice and outgoing, just like not everyone is as tall as a professional basketball player. But here’s the important part, heritability has to do with the origins of these differences. To say that something is heritable is to say that genetic differences play a role in creating observable differences.

Variety in our gene pool matters when we seek to understand why some people can dunk a basketball or compose a sonnet, and why some people persistently break the law. The effects of genetic differences make some people more impulsive and shortsighted than others, some people more healthy or infirm than others, and, despite how uncomfortable it might be to admit, genes also make some folks more likely to break the law than others."


"Most of the evidence about the causes of crime overlooks genetic transmission. Yet, some research has found that once you account for genetic influences on self-control, previously identified social transmission effects (read: parenting) on the child’s self-control become unstable. In other words, when you control for genetic transmission (the alternative explanation that most criminologists overlook), the effect of parenting on self-control diminishes or goes away entirely.

Consider another type of parenting effect — one that shows up in the news frequently — spanking. Not long ago, we examined the relationship between spanking and behavioral problems in children. Once we controlled for genetic transmission, there was no spanking effect in the way that most scholars think about spanking effects. Put another way, our evidence did not support the conclusion that spanking causes behavioral problems in the sense that most psychologists would argue.

The conundrum of heritability transcends parenting. For instance, it’s obvious that crime isn’t randomly distributed across neighborhoods. It seems to be a relatively stable factor that defines an area over many generations. Equally nonrandom, though, is the process by which people sort themselves into neighborhoods. People cluster into areas based on a host of factors, including the primary factor of income. Here’s the kicker, if any of the traits that affect residential choices are heritable and you ignore that influence, your findings regarding the impact of neighborhood factors on crime could be in jeopardy.

A remarkable study in Sweden recently found that highly disadvantaged neighborhoods had more crime. Yet that neighborhood effect disappeared when risk factors concentrated within certain families were taken into account. Once again, social transmission effects weakened (and, in this case disappeared) when other factors like genetic transmission were controlled for. Does this finding guarantee that similar results will emerge in other samples around the world? No. But criminologists rarely consider the possibility that their own studies could be polluted by hidden genetic effects.

The more technical term for this phenomenon is genetic confounding, and there is reason to believe that it is endemic to much of the research coming out of the social sciences in general, and criminology in particular."

Post has attachment
First real-world SHA1 Collision

Post has shared content
Notice and Takedown in the Domain Name System: ICANN's Ambivalent Drift into Online Content Regulation

ICANN is expanding the notion of private law of the internet. The prior lex-ICANNia was (and still is) an expansive new law of trademark-over-everything-else.

Now it is ICANN tacitly standing behind private agreements that will act with the force of worldwide law over assertions of copyright violation.

This is an ill wind, and it will bring no good.

Post has shared content

Post has shared content
And I can't quite believe this, but there's yet another positive story to share today. This is about a paper that came out recently on particle physics. Like an awful lot of papers in particle physics, it proposes an extension to the Standard Model (our best current understanding of the field, which has done remarkably well in predicting an awful lot of things) which can explain a lot of currently open questions about the universe.

However, this paper has some nice features which most papers of this sort don't. There's a sort-of tradition in particle physics (which I'm embarrassed to admit I've participated in) of publishing "pissing on trees" papers: you come up with some theory, show that it's not inconsistent with what we've observed so far about the universe (and it turns out there are an awful lot of things you can do which aren't inconsistent, even once you take the full scientific rigor of professional physicists into account), and publish it as "maybe." This is called "pissing on trees" because if it turns out later that this theory was right, then you've published one of the original papers on it, and a great deal of credit will follow; that is, you're staking out your claim ahead of time, but not really producing anything that valuable, because most of these "maybes" are pretty far-out.

This "SMASH" paper (short for "Standard Model + Axion + Seesaw + Higgs," a short description of the three kinds of extension to the SM it provides) does considerably better, though. With a fairly minimal extension to existing physics (proposing three new families of particle, each of which is considered not-outrageous) they manage to explain a bunch of difficult open problems in physics at once. And rather nicely, the SMASH hypothesis is straightforwardly testable – to the extent that several planned experiments already in the works should be able to say a definitive "yes" or "no" to it within the next decade.

I won't try to give a full explanation of the things it explains, since this gets really technical really fast. The short list is inflation (what force caused the universe to expand really rapidly in its early history, so that its current size is "really big" rather than "about the size of a grapefruit"), reheating (how inflation stops and that energy of expansion somehow gets converted into matter instead of a big, empty universe), dark matter (what is this mysterious substance which appears to form a quarter of the mass of the universe, yet be transparent to light?), baryogenesis (in particular, why is there so much more matter than antimatter in the universe? It's handy for the "not going boom all the time" business, but it's far from obvious why it should be true), and stability (why at daily-life energy scales, certain high-energy properties of physics don't cause Higgs bosons to suddenly become infinitely heavy and attractive or similar weird things which many theories fall victim to).

There are plenty of theories which explain these individually, but SMASH is nice in giving systematic answers to all of them – which makes me far more interested in it than in most papers of this sort.

Of course, it'll take a lot of experiment to see if this goes anywhere, but for once, we actually have an existing experimental roadmap which will answer that. :)

Post has attachment
Removing the last big Hurdle to Silicon Photonics

... except that the center of the modulator — including the waveguide that runs along its top — is undoped. When a voltage is applied, the free carriers don’t collect in the center of the device; instead, they build up at the boundary between the n-type silicon and the undoped silicon. A corresponding positive charge builds up at the boundary with the p-type silicon, producing an electric field, which is what modulates the optical signal.

A conventional silicon modulator is half p-type and half n-type silicon; even the waveguide is split right down the middle. On either side of the waveguide are electrodes, and changing the voltage across the modulator alternately concentrates and dissipates free carriers in the waveguide, to modulate an optical signal passing through.

Post has shared content
Worst case scenario. We are seeing a calamitous failure of the entire democratic experiment. I don’t believe that — not yet. I think the oligarchic putsch has made a big mistake by attacking our professional classes and intel and military officer corps. But I might be wrong. In which case, we will have to study from those who have lived under despots and learn the arts of resistance.

Consider these bits of advice in: “Autocracy: Rules for Survival,” by Masha Gessen.

Rule #1: Believe the autocrat. He means what he says.
Rule #2: Do not be taken in by small signs of normality.
Rule #3: Institutions will not save you.

I am still a sucker for this one. But go read what she says, anyway.

Rule #4: Be outraged. In the face of the impulse to normalize, it is essential to maintain one’s capacity for shock.

Rule #5: Don’t make compromises. Like Ted Cruz, who made the journey from calling Trump “utterly amoral” and a “pathological liar” to endorsing him in late September to praising his win as an “amazing victory for the American worker,” Republican politicians have fallen into line. Democrats in Congress will begin to make the case for cooperation.

History supports this. Democrats always always try to negotiate, while the GOP has become the most tightly disciplined partisan machine in American history, utterly hewing to the “Hastert Rule” (concocted by their former speaker and leader and convicted sexual pervert-predator, Dennis Hastert) to never, ever negotiate in good faith. DP Congresses always meet Republican presidents halfway. (GOP Congresses never do that with Democratic presidents.) But today, “halfway" is still utterly insane treason.

Rule #6: Remember the future.

Says this deep futurist… amen.

Post has shared content
Scientists end long-standing controversy about a ubiquitous reaction involved in catalysts, corrosion, and more
Wait while more posts are being loaded