Is AI Bad for Critical Thinking? It Depends on when You Use It
Researchers at the 2026 CHI conference found that participants who tackled a policy‑making problem on their own before consulting OpenAI's GPT‑4o scored higher on critical‑thinking metrics than those who used the chatbot from the start. The experiment split 393 volunteers into groups with either ample (30 minutes) or limited (10 minutes) time and varied the timing of AI access—early, continuous, late, or none. Those with sufficient time and late‑stage AI assistance produced the strongest essays, while the fastest performance under pressure came from early AI users. The results highlight a trade‑off between speed and independent reasoning.
Machine Learning Streamlines the Complexities of Making Better Proteins
Researchers at UC Berkeley and the Arc Institute unveiled MULTI‑evolve, a machine‑learning framework that predicts how multiple amino‑acid mutations affect protein function in a single experimental round. The workflow first estimates single‑mutation effects, then measures pairwise interactions, and finally trains...
AI Helps Archaeologists Solve a Roman Gaming Mystery
Researchers used artificial intelligence to reverse‑engineer the rules of a Roman board game discovered on a limestone slab in Heerlen, Netherlands. By running virtual players through more than 100 possible rule sets with the Ludii game system, they identified a...
AI Models Spot Deepfake Images, but People Catch Fake Videos
Researchers compared humans and AI in spotting synthetic media, finding AI excels at detecting deepfake images while humans outperform machines on deepfake videos. In tests, AI models achieved up to 97% accuracy on static faces, whereas participants identified only about...
AI Tool AlphaGenome Predicts How One Typo Can Change a Genetic Story
Google DeepMind unveiled AlphaGenome, a deep‑learning model that can process one million DNA bases in a single context, doubling the reach of its predecessor Borzoi. The system predicts eleven genomic functions at single‑base resolution, delivering a 14.7% improvement in gene‑activity...
A Look Under the Hood of DeepSeek’s AI Models Doesn’t Provide All the Answers
DeepSeek’s latest large‑language models have been subjected to a peer‑review analysis that sheds light on the company’s data‑centric training pipeline and architectural tweaks. The study confirms the use of a transformer backbone with sparse‑attention mechanisms and a multilingual corpus exceeding...
Chatbots Spewing Facts, and Falsehoods, Can Sway Voters
Researchers published two peer‑reviewed papers showing that AI chatbots can measurably shift voter preferences by flooding users with facts, even when many of those facts are inaccurate. In a Nature experiment, pro‑Trump bots moved Harris supporters about four points toward...
