Oliver Marsh (tech & politics)
olivermmarsh.bsky.social
Oliver Marsh (tech & politics)
@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/
Olga and Riddle Russia have been doing important analysis for many years, but are now at genuine financial risk after recent developments. If you can help personally, or know of organisations who can help them get back to stability, please do say. I can put you in touch. ridl.io/donate/
February 13, 2025 at 8:08 AM
January 31, 2025 at 10:29 AM
January 31, 2025 at 10:29 AM
January 31, 2025 at 10:29 AM
January 31, 2025 at 10:29 AM
Ranking of tweets with the most community notes:

#1 x.com/realDonaldTr...
January 31, 2025 at 10:29 AM
Same here except this one interestingly... 😎
January 21, 2025 at 11:11 PM
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.
January 19, 2025 at 3:07 PM
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".
January 14, 2025 at 9:52 AM
German politics watchers: this "we said we wouldn't coalition with far-right but turns out alternatives are too complicated" in Austria is what I worry about for next #Bundestagswahl in Germany, but I've been told this is still unlikely from CDU. Any views?

www.zeit.de/politik/ausl...
January 5, 2025 at 6:40 PM
12/ As well as manual coding, we tested systematic at-scale methods (e.g. semantic maps like below). Hard to summarise all we tried, read the full report, but basically: range of possible issues = no clear one-size-fits-all method, different approaches for different issues (i.e. lots of work...)
December 19, 2024 at 6:11 PM
9/ But still much better than Google or OpenAI, who didn't engage. In fact, we got this from OpenAI.

This led to a bigger issue. These bots block automated collection, so we had to use APIs, which may not reflect how "real" bots behave. But we couldn't get support from companies to get round this.
December 19, 2024 at 5:56 PM
/3 We used regional DE elections at start & of Sept, and published interim results inbetween, to track changes and see if/how companies responded.

One big and cool change to see was our interim findings immediately impacting Microsoft Copilot's safeguards - see graph below.
December 19, 2024 at 12:13 PM
@algorithmwatch.bsky.social ending 2024 with a new LLMs & elections report. Also a test of DSA 40.4 Data Access - we even got data!

Key points in thread 👇 (while travelling so maybe installments). Was quite a process, so lots to say!

@clarahelming.bsky.social

algorithmwatch.org/en/llms_stat...
December 19, 2024 at 12:02 PM
Delighted to announce the prizewinner for "most ironic place to be stuck on a stationary train"
December 2, 2024 at 3:35 PM
Some favourite moments from the X DSA Audit Report.

(I actually think the auditor, FTI Consulting seems to have done an ok job with, erm, dodgy material) transparency.x.com/content/dam/...
November 28, 2024 at 2:58 PM