Ranks distinct AI gain and problem claims from the published record. Scores reward impact, independent source strength, scale, confidence, and recency.
Researchers built a multi-stage video-based framework for signalized intersections that combines YOLOv8 detection with OC-SORT tracking to extract vehicle trajectories, uses a dynamic scaling Conflict Region of Interest to reduce data volume, and predicts conflicts with a Spatio-Temporal Graph Attention Network followed by causal forest interpretation. Tested on field video from an intersection in Nanning, China, the ST-GAT model outperformed existing deep learning architectures on training and testing sets.
A peer-reviewed survey study from January to April 2025 asked 52 US dermatology and dermatopathology professionals to rate AI-simplified versions of six fictitious dermatopathology reports. One version used Basic ChatGPT-4.0 with a simple prompt and the other used a custom DermDecoder GPT with a structured 489-word prompt, evaluated for factualness, completeness, and potential harm.
In 76 eyes treated with silicone oil endotamponade for rhegmatogenous retinal detachment, researchers used an automated OCT segmentation tool and a random forest classifier to track retinal layer changes between oil insertion and removal and to predict categorical best-corrected visual acuity change.
Between August 2023 and July 2024, four radiologists interpreted 4577 screening and diagnostic mammograms in a prospective alternating-month design where a commercial AI system was shown or hidden. Reading times from PACS logs, cancer detection rates, and abnormal interpretation rates were compared between AI-assisted and non-AI-assisted months, with reading time analysis restricted to 2917 cases under five minutes.
In a medical school basic science course, researchers deployed a retrieval-augmented teaching assistant that limited large language model outputs to instructor-curated materials across two consecutive cohorts. They tracked when and how students used it and what they asked, finding strategic, context-dependent adoption with heavier use during high-stakes assessments and after hours.
A November 2025 scoping review in BMC Medical Education screened 3,238 records and included 310 publications on AI in undergraduate medical education from 2020 to April 2024, finding 52% of the included literature appeared in just eight months after the prior general review. Reported uses span autonomous tutoring, self-assessment, simulation-based learning, assessment generation and grading, clinical assessment, procedural skills evaluation, and predictive analytics in both basic and clinical courses.
In a study published March 28 2024, researchers examined generative AI tool adoption among software engineers using surveys of 100 engineers and validation with 183 engineers, developing and testing the Human-AI Collaboration and Adaptation Framework.
A peer-reviewed discussion published 5 November 2025 examines whether AI-generated deepfakes used to deceive an enemy during war comply with international humanitarian law. It notes commanders may seek tactical advantage by making opposing forces and civilian populations believe X when Y is true, including fabricated surrender or truce announcements.
ace