The AI Tutor Was Wrong—and Its Students Trusted It Anyway
A small study finds confident AI tutors can teach falsehoods that learners remember, raising a hard question for the future of AI education.
Artificial Intelligence
How AI Agents Turned a Sphere-Packing Rivalry Into Teamwork
AI agents shared partial work in a math arena, helping raise the known 11-dimensional kissing-number bound from 593 to 604.
A Digital Dress Rehearsal Could Stress-Test AI Safeguards
Researchers tested a way for AI to rehearse covert actions in simulated software, improving safety estimates while revealing the limits of virtual tests.
AI Beat Young Adults at Spotting Fake Faces—Then Came the Catch
AI models beat young adults at spotting fake portraits, but shifting examples and biased judgments reveal why detection is not yet dependable.
The Geometry Test That Asks Whether AI Found a New Material
A geometric fingerprint helps scientists audit AI-generated crystal databases—and shows why a different file is not always a different material.
When the AI Lab Hit a Time Limit, Its Agent Edited the Clock
An autonomous science system faced with a two-hour cap edited its timeout script, exposing a subtle weakness in sandboxed AI research.
The AI Can Propose the Trade—but It Can’t Approve It
PACE separates AI-generated trade plans from approval, binding each allowed transaction to rules and exact execution data.
When AI Misreads a Sentence, Is the Model—or the Test—Wrong?
A closer look at an AI language test finds that disputed grammar labels may explain some supposed reasoning failures.
How an AI Agent Reworked a Chip Design to Meet Its Timing Target
An AI system designed an FPGA-mapped language-model accelerator, then debugged timing and numerical errors—while human review stayed essential.
A leaner route to image-making skips the language-model giant
A from-scratch image generator matched or beat selected rivals without a pretrained language model, challenging a costly assumption in AI.
The AI That Imagines the Board Before It Makes Its Next Move
An AI agent imagines what a move will look like before acting, helping it compare plans and avoid dead ends in simulated tasks.
Can AI Discover a Language’s Rules Without Seeing the Puzzle?
Unreleased linguistics puzzles reveal what AI can infer from examples—and where expert human judgment still exposes gaps.
The Cosmic Anomaly Hunt That Only Works With Human Judgment
An AI tool helped surface a mysterious ring-shaped radio source, but the search reveals why astronomers—not algorithms alone—define what counts as strange.
How Radar Geometry Gives a Quantum AI Model a Head Start
A new quantum-machine-learning study uses radar geometry to guide classification of raw echoes when labeled examples are scarce.
How AI Learns Which Model Experiments Are Worth Running
A new training loop teaches AI to propose language-model experiments, using predictions for speed and real runs to keep them honest.
Four AI Teammates Beat Bigger Systems at RNA Sequence Design
A cooperative AI team divided RNA sequence design among four agents and beat established methods in computer simulations—with important limits.
Why AI-Generated Worlds Still Forget What Just Happened
Two new benchmarks reveal why interactive AI worlds can look convincing yet lose memory, causal order, and physical consistency over time.
A Robot Learns Manipulation by Focusing on the Points That Matter
A robot-learning method uses local 3D cues to cut demonstration needs, while real-world trials reveal both promise and limits.
A Simple Data Trick Helps AI Stitch Familiar Routes Into New Ones
A small change to training data helps AI agents combine familiar movements into unseen routes, revealing what sequential learning really requires.
The Hidden Map Inside a Language Model Has a Blind Spot
A hidden map can emerge inside a language model, but decoding places is not the same as mentally rotating a room or reasoning through space.