Jonathan Stray

@jonathanstray

Working on better personalized news and information at . Previously: computational journalism , editor , made

San Francisco
jonathanstray.com
Joined May 2008
960 Photos and videos Photos and videos

Tweets

You blocked @jonathanstray

Are you sure you want to view these Tweets? Viewing Tweets won't unblock @jonathanstray

  1. Pinned Tweet
    Jan 21

    If "engagement" shouldn't drive media algorithms, what should? held a workshop with 40 recommender system professionals, researchers, and critics and here are our answers.

    1 reply 10 retweets 33 likes
    Undo
  2. Jul 30

    A better headline might have been: "CDC study shows vaccines don't protect against Delta as well as earlier variants, underscoring the importance of herd immunity."

    1 reply 2 retweets 12 likes
    Show this thread
    Show this thread
    Undo
  3. Jul 30

    If you want to get serious about this, you could compute the risk ratio between vaxxed and unvaxxed, using a base rate of 69% vaxxed for the county (as the CDC paper notes). I've seen similar cases of failure to use a risk ratio in journalism before

    2 replies 2 retweets 8 likes
    Show this thread
    Show this thread
    Undo
  4. Jul 30

    Part of the journalistic challenge here is there are two key messages to get across, that seem to pull in opposite directions: - you can still get infected even if vaxxed - but vaccines really cut down the risk Together, this is the reason why everyone needs to get vaxxed.

    1 reply 1 retweet 13 likes
    Show this thread
    Show this thread
    Undo
  5. Jul 30

    If this isn't clear to you, consider: when 100% of the population is vaccinated, then 100% of the COVID cases will be in vaccinated people. There is no information about the effectiveness of the vaccines in this headline, even though it sounds like there is.

    2 replies 6 retweets 36 likes
    Show this thread
    Show this thread
    Undo
  6. Jul 30

    This is an awful, awful, headline. I've been teaching stats to journalists for years but I never thought it would be life or death (base rate fallacy). Should have gone with "Roughly 97% of new hospitalizations and 99.5% of deaths in the U.S. are among unvaccinated individuals"

    CDC says 74% of people infected in Massachusetts Covid outbreak were fully vaccinated
    9 replies 75 retweets 207 likes
    Show this thread
    Show this thread
    Undo
  7. Jul 30

    This is one of those rare cases where an understanding of Bayes' Theorem is actually critical. "% infected people who were vaxxed" is really, really not the same as "% of vaxxed people who were infected." If 90% are vaxxed, most infections will be vaxxed.

    CDC says 74% of people infected in Massachusetts Covid outbreak were fully vaccinated
    1 reply 5 retweets 8 likes
    Undo
  8. Jul 28

    Platform recommender engineer told me today: yes, we could turn up the diversity so no one's feed gets too specific. But then a lot of people would miss out on finding all the other folks who are into whatever their thing is, and people get a lot of joy and connection from that.

    1 reply 2 retweets 9 likes
    Undo
  9. Jul 28

    Sometimes people are like "why is polarization bad?" and I'm like, well, I reeallly don't want to go any further up this escalation curve.

    1 reply 8 retweets 48 likes
    Undo
  10. Retweeted
    Jul 27

    Still, these data paint a starkly different picture from the CrowdTangle dashboard that Kevin Roose regularly tweets. See

    This is the $ graph. In 2018 84% of article shares on FB were from credible news domains, while 89% of views were from credible news domains (though I'm sure people will quibble w/ sites like the Daily Wire being categorized as credible)
    2 replies 1 retweet 6 likes
    Show this thread
    Show this thread
    Undo
  11. Jul 28

    Remember when I said that the White House might be the worst possible messenger for trying to get people vaccinated? This backfire effect has been confirmed.

    Study: "Elite party cues increase vaccination intentions among Republicans" by , et al. Yes, vaccines are now inescapably political (big ). We need Republican voices (PLEASE step up!) in order to encourage hesitant conservatives.
    Undo
  12. Jul 27

    We just aren’t there yet with basic recommender controls.

    I’m not interested in regular sports, yet based on the “For You” trends, the Twitter algorithm thought I was a sports fanatic. Every trending topic would be sports, all the time. I would use the drop-down to say “I’m not interested in this” but it never made a difference…
    Show this thread
    1 retweet 6 likes
    Undo
  13. Retweeted
    Jul 25

    I am awestruck at this combo of fluent data analytics for non-coders plus training for data journalists. Kudos , , for advancing what I first saw in DabbleDB () and OpenRefine ().

    4 retweets 5 likes
    Undo
  14. Jul 24

    I see this correlation too. This thread picks apart potential reasons for that, but more importantly how this cultural fact might bias guesses about what types of AI social impacts will prove most significant.

    To be more precise, I think there is a strong correlation in AI ethics/policy world between (A) thinking that present/past AI achievements are overhyped and future progress is likely to be slow, and (B) left leaning political views and skepticism re: big tech companies/elites.
    Show this thread
    2 likes
    Undo
  15. Retweeted
    Jul 23

    I'm going to repeat this -- please read this whole thread, which is one of the smarter things I've read on this issue. Whether respect is useful as a strategy for deniers depends almost entirely on your location in the network.

    This is why the claim "yOu MusT reSPekT vAccInE dEnieRs" is BS: vaccine deniers don't crave respect from *everyone.* Randos offering respect won't change a thing. The only people who can sway these folks are members of their chosen reference groups: maybe family, church, &. 24/x
    Show this thread
    1 reply 58 retweets 112 likes
    Show this thread
    Show this thread
    Undo
  16. Jul 23

    The tweet it composed for me was pretty good, actually. Full points for self-reference.

    2 likes
    Undo
  17. Retweeted
    Jul 17

    Good. I’m sick and tired of classic artists being exempt from the attention economy. They should feel the dehumanizing pressure of engagement metrics just like the rest of us.

    Using cameras to monitor people looking at artwork, researchers hope that the data can help define "attraction value" for specific works of art, leading to changes in museum and gallery layout and exhibit scheduling
    49 replies 1,049 retweets 5,057 likes
    Undo
  18. Retweeted
    Jul 22

    Instagram, which is owned by Facebook, is now targeting me with ads about the awesomeness of targeted ads

    2 replies 3 retweets 1 like
    Show this thread
    Show this thread
    Undo
  19. Jul 22

    My takeaway: this is a very cool algo auditing method. But it’s recsys 101 stuff. The hard question is not whether a recsys responds to your interests — it does! — but what's happening with the humans, and what kinds of media diets are healthy. Bot audits can’t answer that. /FIN

    1 reply 8 likes
    Show this thread
    Show this thread
    Undo
  20. Jul 22

    To close, a word on transparency. I’d love the reporters to publish their code so we can see exactly what they did, and how to build on it. The deeper problem is it should be possible for outsiders to run audits on real platforms. Perhaps like this. 15/

    1 reply 6 likes
    Show this thread
    Show this thread
    Undo
  21. Jul 22

    The reporters also found it surprising that only one signal — watch time — was enough to establish each bot’s interests. I don’t find this surprising. Both theory and practice show that a single reward signal can produce incredibly complex behavior. 14/

    2 replies 1 retweet 4 likes
    Show this thread
    Show this thread
    Undo

@jonathanstray hasn't Tweeted yet.

Loading seems to be taking a while.

Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.

    You may also like

    ·