AI Disease-Prediction Models Raise Reliability Concerns
Researchers, led by Adrian Barnett from the Queensland University of Technology, have identified that dozens of AI disease-prediction models for stroke and diabetes were trained on questionable data sets, raising concerns about their reliability in clinical settings. An analysis revealed that 124 peer-reviewed papers utilized these data sets, which exhibited irregularities suggesting potential fabrication, with at least two models already in use in hospitals in Indonesia and Spain. Public-health researcher Soumyadeep Bhaumik emphasized that models based on unverified data are unreliable and could lead to inappropriate clinical decisions.
Physicists resolve proton size puzzle with new measurements
Physicists have proposed a resolution to the long-standing "proton size puzzle," which has seen conflicting measurements of the proton's charge radius over the past 15 years. Recent ultra-precise hydrogen measurements indicate a smaller proton core, aligning results that were previously inconsistent and suggesting that the proton's radius is approximately 0.876 femtometers. This development may restore confidence in using hydrogen spectroscopy for fundamental physics tests and refocus research efforts on remaining anomalies.
Study finds optimal AI use enhances critical thinking
Researchers led by Mina Lee from the University of Chicago found that individuals who consulted an AI chatbot later in the problem-solving process performed better on critical thinking tasks than those who used it from the start. The study, presented at the 2026 CHI conference in Barcelona, involved 393 participants who were assigned to various groups based on their access to the chatbot and the time allotted for the task. Results indicated that while early AI use could enhance performance under tight deadlines, waiting to engage the AI allowed for deeper reasoning and better retention of information.
