A new approach uses diff-based analysis to track whether individual lines of text were written by a human or generated by an AI, specifically in contexts where agents iteratively edit documents. This provides granular provenance, making it easier to audit mixed human-AI contributions. The technique matters for maintaining transparency and accountability in AI-assisted writing workflows.

A study found that taxi and ambulance drivers die from Alzheimer's disease at significantly lower rates than the general population. The finding suggests that the constant spatial navigation and memory demands of these jobs may help build cognitive resilience against Alzheimer's. This adds to evidence that mentally active work could protect brain health in old age.

A built-in Windows 11 Weather app has been found to use more than 1 GB of memory, which is unexpectedly high for a simple utility. This issue can degrade system performance on PCs with limited RAM. Users are advised to monitor the app or disable it if unnecessary.

Historian Jill Lepore argues that Silicon Valley misinterprets science fiction as a guide for the future, leading to flawed technological and political visions. She warns this mindset undermines democratic institutions and civic life. The critique adds to growing scrutiny of tech leaders' influence on society.

A 1998 article by Tim Berners-Lee, titled "Cool URIs Don't Change," is highlighted, arguing that web addresses should remain stable and permanent. He emphasizes that changing URIs breaks hyperlinks and erodes trust in the web. This principle remains a foundational guideline for web design and information architecture.

A developer has created an interactive globe that projects modern city locations onto Earth's tectonic configurations in the past and future. Users can visualize how continents have shifted over millions of years and will continue to move. This tool makes deep geological time and plate tectonics more accessible and engaging.

TechCrunch reports that AI safety testing is introducing new dangers, as the tests themselves can be exploited, produce misleading results, or cause unintended harms. This matters because relying on flawed safety evaluations could give a false sense of security as AI systems become more powerful.

A new mathematical article examines the recently discovered counterexample to the HRT conjecture in time-frequency analysis, which challenges a long-standing assumption about the linear independence of Gabor systems. The post breaks down the counterexample's structure and implications for signal processing and harmonic analysis.

Researchers have created an adversarial pattern that, when worn or printed, tricks AI-powered surveillance systems into failing to detect a person. The pattern exploits vulnerabilities in object detection models, allowing wearers to evade camera recognition. This development raises significant privacy and security questions about AI surveillance.