@_FelixSimon_
🇩🇪Account based in
About this account
- Account based in
- Germany
- Connected via
- Germany App Store
Account-level information from X, not a live location or the device used for a specific post.
Research Fellow in AI & Information @risj_oxford | Research Associate & PhD @oiioxford | Societal impacts of AI, news, (mis)info, democracy | My views etc…
Oxford & the bird app 🫡
Joined September 2012
- Tweets11.3K
- Following1.2K
- Followers4.9K
- Likes13.4K
Pinned Tweet
How harmful is GenAI around elections? Will it trigger a misinformation apocalypse and upend elections?
I am happy to finally be able to share @Sacha_Altay’s & my answers to these and other questions on which we have been working for a year and which is out via @knightcolumbia.
Just got told by someone far smarter than me that one way you can tell Anthropic is really successful in some circles (i.e. white collar workers) is that Claude has become a verb (“to claude”) much in the same way as Google/“googling” before.
Felix M. Simon retweeted
New in @PNASNews:
In two large experiments (~18k obs from ~14.8k people), we find that brief conversations with AI can get people to take real political action - incl. signing real petitions and giving money to charity.
Surprisingly, the persuasion strategy that worked best for changing people's attitudes - packing convos with info - was the least effective at changing behavior!
🧵
How do people use AI, what do they think about its impact on society, and its use around information, news and journalism specifically? Not long now until we have answers for 2026 for you – for nine countries, no less!
For last year's edition, see here. buff.ly/xY53IFi
Felix M. Simon retweeted
Randomizing uptake of NPR's @UpFirst podcast for 5 days increases news consumption & knowledge of events (esp. among more right-leaning voters), trust in NPR, etc but not trust in the news more generally; no effect on attitudes or political beliefs. An @andyguess et al. joint
Felix M. Simon retweeted
True re: attention! And glad Anthropic is starting to share this data. But I struggle to interpret these numbers:
-IIUC, they're reporting share of conversations, not users. If a small % of heavy users have lots of other conversations (e.g. about coding), which seems plausible to me, then 0.51% of convos could mask a much higher share of users (the election-relevant quantity) asking about the election
-6 weeks before a midterm is still before most voters tune in
Hope a fuller release has per-user data across the full cycle!
The attention constraint on AI persuasion is fundamental, as we argued in this chapter preprints.apsanet.org/engage…
Just 0.51% of Claude conversations are about elections or voting; at most 0.01% about who to vote for
"It feels like the talk about hallucinations has somewhat died down" – this observation from my friend Kirsten Eddy who hosted me at Pew Research today for a talk got me thinking. At least to me, it feels like she is right.
As of late I've seen fewer and fewer discussions of something that was an incredibly salient issue when genAI kicked off. Models getting better will have played a part, perhaps also people getting used to AI.
Felix M. Simon retweeted
Even the most "AGI-pilled" people now typically acknowledge that merely making AI systems generally "smarter" won't straightforwardly translate into society-wide benefits – creating material abundance, curing diseases, improving governance – because in many areas we're not just bottlenecked by a lack of raw intelligence or explicit intellectual knowledge.
My impression is that this lesson isn't internalised to the same degree when it comes to how more intelligent machines could cause societal harms. If you need far more than raw general intelligence to cure cancer, you also need far more to create effective propaganda, let alone take over the world or eliminate the human species.
When I wrote about the limits of AI persuasion in the wild for Transformer recently, I didn't think of warning labels/or literacy treatments as a limiting factor – but seems like this cuts persuasive effects (see new WP by @DG_Rand & @ROrchinik).
arxiv.org/abs/2609.16432
Can’t help but wonder how the Navier-Stokes controversy will/won’t impact how researchers are using frontier firms’ models or science environments.
ft.com/content/4d6d28a0-4b12…
But I wonder if mathematicians, biologists, or STEM folks (especially those working on research that could lead to commercial breakthroughs) will sleep soundly going forward, let alone different companies using these systems.
While I cannot be at ECREA this year to co-present, please check out our AI work presented by great colleagues Richard Fletcher, Amy Ross Arguedas and Waqas Ejaz.
Today at 12:30, Richard will present our 6-country findings on whether AI-rewritten news make people engage more and are easier to understand and more interesting (yes on all counts).