Arvind NarayananOvjeren akaunt

@random_walker

Princeton prof. I use Twitter to share my research & commentary on surveillance capitalism, infosec, cryptocurrencies, AI ethics, tech policy, & academic life.

Vrijeme pridruživanja: prosinac 2007.

Medijski sadržaj

  1. Unfortunately, all five carriers used authentication methods that are considered insecure in the computer security community. Taken together, these findings help explain why SIM swaps have been such a persistent problem. More details in our paper:

    Prikaži ovu nit
  2. Privacy advocates have long said it should be treated as a human right. Finally we have an economist supporting this view: Alessandro Acquisti (), who did many of the pioneering studies of how framing affects privacy. Essay by .

    Prikaži ovu nit
  3. It's possible that reputation management firms are generating RTBF requests on behalf of the powerful and privileged. The good news is that only 16% of requests are from high-volume requesters. But ideally Google should release more data about that 16%.

    Prikaži ovu nit
  4. There's a fairly uniform rate of RTBF utilization among EU countries. The small differences can be explained by social media penetration & differences in journalistic practices w.r.t. reporting private info (the paper admits this despite playing up the differences b/w countries).

    Prikaži ovu nit
  5. In the majority of cases it's clear there's a legitimate intent behind the request, such as delisting legal history. Many of the requests are for "directory services" aka skeezy data aggregators. The big drop mid-2018 is because the GDPR put a bunch of them out of business. Good!

    Prikaži ovu nit
  6. Despite fears that people might overwhelm search engines with bogus requests, Google's rate is only 50K/month, which is a completely reasonable privacy burden for a company of that scale. About 45% were successful, which suggests that it's not dominated by bogus requests.

    Prikaži ovu nit
  7. For three years I had a great time helping and others run ConPro. It is a true workshop, aimed at improving in-progress research. It is back this year with and promises to be even better. Submit a paper or talk proposal by Jan 23.

  8. Odgovor korisniku/ci
  9. Also worth noting: –The FTC seems to be OK with this half-assed measure –Reuters' headline parrots Facebook's spin: "Facebook separates security tool from friend suggestions, citing privacy overhaul" –US privacy enforcement relies on companies pinky swearing. No external audits:

    Prikaži ovu nit
  10. Update: Facebook now promises to stop misusing phone numbers intended for 2-factor authentication, by which of course I mean that Facebook has simply added new hoops for users to jump through if they don't want their phone numbers misused. HT

    Prikaži ovu nit
  11. Update: despite previously telling the editor why I think Elsevier is a menace to society, I still get review requests from the same Elsevier journal. Now there's an extra disincentive—they could trivially fix this bug that leaks reviewer identities, but they've chosen not to. 🤦🏽‍♂️

  12. The original filter bubble was a hall with terrible acoustics, ten times as long as it was wide. Like-minded delegates congregated with each other and literally couldn’t hear the opposing side.

    Prikaži ovu nit
  13. Technologists who think platforms for online speech are neutral and don’t have politics should remember that the left-right political spectrum dates to the French revolution, wherein the geometry of the meeting hall helped fragment the debate and polarize the members' positions.

    Prikaži ovu nit
  14. The authors ( ) build on their previous work where they reverse-engineered FB's algorithms using really neat techniques like finding that ad images invisible to humans still result in skewed audiences

    Prikaži ovu nit
  15. Interesting way to quantify the skew: attempting to undo the effect of Facebook's targeting is up to 50% more expensive for the advertiser.

    Prikaži ovu nit
  16. Facebook's algorithms skew the delivery of political ads towards partisan audiences by inferring users' political affiliations and automatically analyzing ad content. New paper by researchers at Northeastern, USC, and Upturn:

    Prikaži ovu nit
  17. A bit of history: we proposed perceptual ad blocking in 2017. Facebook claimed back then that its ads were unblockable. But we released an extension that successfully blocked FB ads based on the legally mandated disclosures in those ads.

    Prikaži ovu nit
  18. AdBlock Plus now has perceptual ad blocking, which identifies ads based on what they look like instead of where they're served from. Visually detecting ads is a more scalable method than applying tens of thousands of (manually created!) filter rules.

    Prikaži ovu nit
  19. Fun fact: sorting an array using "Math.random() - 0.5" as the comparison function doesn't produce a random shuffle. This exact bug led Microsoft to favor Chrome in its UI for choosing a browser in response to EU antitrust investigations back in 2010 😲

  20. Word embeddings are not the right place to locate fairness interventions. This point is also made in the Fairness and Machine Learning textbook by , and me.

Čini se da učitavanje traje već neko vrijeme.

Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.

    Možda bi vam se svidjelo i ovo:

    ·