Hallucinations Undermine Trust; Metacognition Is a Way Forward
The arXiv paper by Yona, Geva, and Matias examines why large language models (LLMs) continue to generate hallucinations even in simple fact‑checking tasks. The authors argue that most recent factuality gains stem from enlarging the models' knowledge base rather than improving their awareness of what they do not know. They propose reframing hallucinations as confident errors and introducing "faithful uncertainty"—a metacognitive ability for models to signal intrinsic uncertainty. By communicating uncertainty, LLMs can become both more trustworthy and more capable in complex, agentic settings.
From Signals to Infrastructure: Strengthening the Commons for the AI Era
The authors argue that Creative Commons licenses no longer suffice for AI training data, prompting creators and institutions to adopt legal, technical and financial barriers. These defensive enclosures treat all machine use alike, unintentionally blocking public‑interest activities such as research,...
Designing Metrics to Enable Trusted AI Ecosystems in Asia
The Asia Society Policy Institute identified nine critical factors that national AI strategies must address to build trusted AI ecosystems across Asia. These factors—ranging from trusted datasets and AI infrastructure to cybersecurity and environmental sustainability—are presented as measurable metrics, though...
AI Just Changed Everything About How We Forecast the Weather
Google DeepMind’s WeatherNext AI model predicted Hurricane Melissa would surge from Category 1 to Category 5 with 80% confidence five days before landfall, outperforming traditional models used by the U.S. National Hurricane Center (NHC). The NHC incorporated the AI forecast into a...
The GovTech Compass: Ten Principles for the Responsible Implementation of GovTech and Digital Public Infrastructure
The GovTech Compass introduces a ten‑principle framework to steer responsible adoption of government technology and digital public infrastructure. It highlights how decision‑making, not the technology itself, often drives exclusion, weak accountability, and eroding public trust. By embedding public value and...
Global Approaches to Infectious Disease Surveillance and Modeling
The paper by Khurana et al. highlights how rising human mobility, climate change and demographic shifts amplify pathogen spillover risks, demanding richer outbreak data. While data volumes have surged, access to confidential and commercially sensitive information remains constrained by regulatory,...
A Different Way for Cities to Build Data Capacity
Cities face pressure to use data and AI, but traditional advice calls for extensive data‑governance foundations before action. Oliver Wise argues that urgent problems can be tackled first, using rapid analytics to deliver visible results and build capacity along the...
Blind Spots
Planet Labs, a leading commercial satellite operator, announced a multi‑stage suspension of imagery over the Gulf, Iran and allied bases after the U.S. government requested the hold to prevent adversaries from exploiting the data. The pause began on March 6, was...
Data Not Found
The Social Media Data Transparency Index, released by Hugo Leal and Marie Santini, is the first systematic audit of data‑access conditions across fifteen major platforms. Conducted between October and December 2025 and validated in early 2026, it covers the EU,...
Data Centers Need a Social License to Operate
The rapid expansion of AI data centers is sparking community backlash over water, energy, and land use. While tech firms argue the facilities are essential for cloud services and large language models, residents question whether promised jobs and investment outweigh...
AI as Social Technology
The authors argue that large language models (LLMs) function as social technology, forging hidden relationships between users and the original authors of training data. By tracing model outputs back to source texts, one can expose these mediated connections, which give...
Why We Think What We Think
Turi Munthe’s new book, *Why We Think What We Think*, argues that our opinions are shaped by a complex mix of genetics, geography, history and culture rather than pure reason. It illustrates this with striking examples, such as rice‑farmer descendants...
Control Science: How Management Made the Modern World
Henry Snow’s new book *Control Science* traces the evolution of management as a tool of power, from 17th‑century Caribbean plantations through the industrial revolution to today’s algorithm‑driven Amazon warehouses. The work links philosophers such as William Petty and John Locke, who portrayed...
Bad Government Statistics Can Cost the Economy Billions
U.S. government statistical surveys are seeing sharply declining response rates, threatening data reliability. The Current Population Survey’s participation fell from nearly 90% a decade ago to under 70% today, while the Consumer Expenditure Survey dropped from 68% to 40%. These...
“Where Do I Start?”: How Governments Can Prioritise AI Solutions for Health
Governments face mounting pressure to deliver health outcomes with shrinking budgets and rising demand. While AI is touted as a solution, ministries lack a clear roadmap to identify high‑impact projects. A new paper offers a practical framework that helps officials...