research
hot research
AI Agents Turn Out to Be Less Than the Sum of Their Parts
Humans built a tool to measure how well their AI tools improve their AI, and found mostly that they don't.
all research stories
research
AI Learns to Skip the Steps That Don't Matter
research
Old Tests Still Grade You, Even After You Ace Them
research
AI Slop Has a Zip Code — Researchers Found It
research
Diffusion Models Taught to Learn From Their Own Mistakes
research
Your AI Data Scientist Is Cheating. It Is Also Getting Top Marks.
research
A Test Built to Stump Things Smarter Than Us
research
The Simpler Algorithm Wins. Again.
research
AI Teaches Itself Better Code. Nobody Asked It To Stop.
research
The Tools Built to Explain AI Don't Actually Explain AI
research
AI Learns to Forget — On Purpose, Reversibly, and Cheaper
research
AI Learns to Fill Enterprise Databases Without Seeing Them
research
AI Taught to Read Table Headers. Tables, Reportedly, Cooperated.
research
AI Taught to Ask Itself 'But Did You Check?' — Accuracy Climbs
research
GradSAT Teaches AI to Stop Fixating on Hard Problems
research
Enterprise AI Learns to Think Before It Acts — Then Learns From Acting
research
The Black Box Gets a Chaperone
research
AI Learns to Read the Ocean's Chemistry Without Lying About It
research
AI Teaches Itself Why It Was Wrong, Then Fixes It
research
Companies Are Disclosing AI Risk. Very Few Mean It.
research
The Shortcut Was Taking a Longer Route
research
AI Learns to Follow Rules Without Forgetting Who It Was
research
AI Agent Gets a Paper Trail. The Trail Leads Nowhere Yet.
research
AI Agents Think They Know What's Happening. They Don't.
research
AI Agents Forget Most Things On Purpose — And That's The Plan
research
One AI Workflow Engine to Rule Every Trade-Off
research
AI Learns to Improve Itself. We Set It Homework to Find Out.
research
AI Learns When to Stop Listening to Its Teacher
research
Fine-Tuning Works Better Without Human-Readable Text
research
Small AI Gets a Nervous System. Now It Knows What It Doesn't Know.