olivermmarsh.bsky.social
Head of Tech Research, @algorithmwatch.bsky.social. Will bridge worlds of technology & society for money. Counter-online-harms, government&tech, regulation, data access, etc. Based in Berlin. https://www.oliver-marsh.com/
100 posts
257 followers
212 following
Regular Contributor
Active Commenter
comment in response to
post
Also: Something quite dodgy seems to have happened with the removal of an anti-Musk tweet. It wasn't a systematic problem, but still weird. Full story in section 4 of the report. Again, the blurring of personal and political in Musk's involvement with X is not great.
comment in response to
post
So the "Musk Effect" we see is: in an important German election, engagement on X is skewed by the personal politics of one hyper-partisan and often inaccurate US Tech Billionaire. That doesn't seem to show a helpful "public square" for discussions of elections.
comment in response to
post
The huge gains in engagement made by the AfD are largely Alice Weidel's English-language content about Musk being boosted by Musk. And the top content anti-AfD organisations was also about Musk.
comment in response to
post
Nor do we see clear evidence that X is tweaking algorithms to impact certain parties (though, as usual, this research is made harder by lack of transparency from platforms 😔).
comment in response to
post
We focused on engagement on posts by German politicians. This adds to other research on the campaign, which does suggest the existence of various influence operations and bot-driven campaigns. But based on our research, this does not seem to be changing engagement on politicians more generally.
comment in response to
post
# 5 x.com/JoeBiden/sta...
comment in response to
post
# 4 x.com/elonmusk/sta...
comment in response to
post
# 3 x.com/elonmusk/sta...
comment in response to
post
#2 x.com/TuckerCarlso...
comment in response to
post
Adding to doc, thanks
comment in response to
post
Same here except this one interestingly... 😎
comment in response to
post
been compiling a list here (from Germany) bsky.app/profile/oliv...
comment in response to
post
these are just searches yes, longer list her - some are so specific I'd doubt Meta was overwhelmed? bsky.app/profile/elio...
comment in response to
post
#project2025 blocked, #hegseth #gaetz #greenland #mexico #panamacanal all fine
comment in response to
post
#dei #musk #gender #biden #lgbt all fine
comment in response to
post
#jan6th and #january6th blocked #capitol not
(I'm in Germany btw)
comment in response to
post
Meta's transparency reports: transparency.meta.com/reports/regu...
comment in response to
post
But the key point: it seems a small number of fact-checks can multiply up into a much larger number of posts affected.
May still be a tiny % of overall posts, maybe even of posts which _should_ be fact-checked. But it doesn't seem negligible.
(But pls point out things I've missed / misunderstood!)
comment in response to
post
But FB DSA Transparency report says ~19m posts demoted* as "fact-checked misinformation", mostly automated but ~11k not, from 01.04.24-30.09.24 i.e. ~100k auto and ~63 non-auto per day. EU also seems to have ~3x as many fact-checkers as USA.
* Not removed, as "censorship" narrative may suggest.
comment in response to
post
I've also been looking at this plus active "rankers" of notes, to explore if "gaming" the system can be spotted, but that'll take longer. Hope useful in meantime!
comment in response to
post
By going into the details of notes you can see note-writers accounts. By playing around with downloaded data you find very active note-writers. Some of them have specific interests, e.g. this one looks for scams x.com/i/communityn... . May be useful if you share their interests.
comment in response to
post
As per trad social media research, you then have content and accounts which could be used as a "seed" for further monitoring. E.g. first one I tried (below) was a well-known dodgy account, but the other (above) was new to me.
x.com/i/birdwatch/...
comment in response to
post
Then - and this is a cool hack - you put the tweet ID into this URL : "https://x.com/i/birdwatch/t/XXX" and you see the original tweet. E.g. x.com/i/birdwatch/...
(in general if you have ANY tweet ID, you can try and if it's been noted you can see the OG tweet, like old API days...).
comment in response to
post
Just from some quick word searches you can find topics you're interested in. Here, for instance, are text notes mentioning all of "AI", "generate", and "biden".
comment in response to
post
But all Notes data can be found here communitynotes.x.com/guide/en/und... with the data you can see all the notes and query the text. (They're big files so I'd recommend not trying this in Excel).
comment in response to
post
I've already said that notes marked "helpful" can be found here. communitynotes.x.com/guide/en/und...
However (i) that's a minority of notes (ii) the text of the notes can't be queried. But much better is....
comment in response to
post
why yes I am having a fun saturday eve, why ever do you ask?
comment in response to
post
I wonder if the 2 could combine to greater than sum of parts. E.g. When I look at Community Notes data it seemed dominated by polarising topics, plus scams. I'd be interested to know if (i) this is true (ii) if that's a problem and (iii) if so, could fact-checkers fill any important gaps
comment in response to
post
q's that interest me (i) will Meta make Community Notes data available, including ratings, like X and (ii) why not use BOTH Fact Checkers and Community Notes (iii) topics: when I look at X notes data it seems they "follow the crowd" to high-profile topics, is this true & how do factcheckers compare?
comment in response to
post
"Helsinki" in scare quotes gave me a much needed giggle, thanks!