Screenshot of an actual Truth Social post from POTUS

Bunch of papers and reports worth listing:

News Integrity in AI Assistants (BBC)

• Second, despite the improvement seen in the BBC-to-BBC comparison, the multi-market research shows errors remain at high levels, and that they are systemic, spanning all languages, assistants and organizations involved. Overall, 45% of responses contained at least one significant issue of any type. Sourcing is the single biggest cause of significant issues (31%). Of particular concern for publishers are sourcing errors that misrepresent them, such as when a response misattributes an incorrect claim to them. Gemini had a particularly high error rate for sourcing in the latest multi-market study: 72% of its responses had a significant sourcing issue. All other assistants were below 25%.
• And yet, many people do trust AI assistants to be accurate. separate BBC research published at the same time as this report shows that just over a third of UK adults say they completely trust AI to produce accurate summaries of information. This rises to almost half of under 35s. That misplaced confidence raises the stakes when assistants are getting the basics wrong. These shortcomings also carry broader consequences: 42% of adults say they would trust an original news source less if an AI news summary contained errors, and audiences hold both AI providers and news brands responsible when they encounter errors. The reputational risk for media companies is great, even when the AI assistant alone is to blame for the error.
• If AI assistants are not yet a reliable way to access the news, but many consumers trust them to be accurate, we have a problem. This is exacerbated by AI assistants and answer-first experiences reducing traffic to trusted publishers.

Contextualizing Misinformation: A User-Centric Approach to Linguistic and Topical Patterns in News Consumption | Proceedings of the ACM on Human-Computer Interaction

• Exposure to misinformation poses significant challenges to democratic processes and public health, particularly during critical events like elections. This study adopts a user-centric approach to analyze the linguistic features of misinformation actually consumed by individuals during web browsing. Using data from a nationally representative panel of 1,240 American adults and their web-browsing data (21M URL visits) during the 2020 U.S. Presidential Election, we examine linguistic and topical differences in the content of 91K unique misinformation and hard news webpages by utilizing natural language processing techniques and Large Language Models. We find that misinformation consumed by users is generally easier to read, exhibits higher negative sentiment, and employs more moral language than hard news. We also find significant linguistic variations across topics–misinformation can be diverse and vary in linguistic features depending on the subject matter. We also identify heterogeneity across key user characteristics: older adults consume more misinformation about COVID-19 and health, with content showing more negative sentiment and fewer moral terms than expected. Republicans engage with misinformation characterized by more negative sentiment and higher moral language, focusing less on health topics and more on social and political issues. These results highlight the importance of a user-centric approach and suggest that interventions to combat misinformation should be tailored to specific topics and user characteristics for greater effectiveness.

Veiled Power: How Rosenwald Teachers Quietly Shaped the Civil Rights Movement

• What precipitates the collapse of seemingly durable social orders like Jim Crow? During the 1920s, approximately 5,000 “Rosenwald Schools” were built across the rural South through a partnership between philanthropist Julius Rosenwald and Black communities who raised matching funds, donated land, and petitioned local governments. Local elites saw vocational training that would preserve the racial order. We argue Black educators used this accommodationist cover to build veiled capacity: organizational infrastructure for collective action behind a veil of compliance. Counties with more Rosenwald Schools show greater civil rights protest in the 1960s. Mediation analysis reveals that pre-existing social capital predicted protest through Rosenwald teacher placements, not overall Black enrollment. Instrumental variable models suggest the effect is not driven by community selection. Moving from no Rosenwald teachers to the 75th percentile predicts 45% more protest. The political effects of education may depend less on what elites intend than on what educators build where elites cannot see.