Before drafting with AI, agree on what the words mean
The paper uses examples to show how people and AI can mean different things by the same word. Compare two meanings of a key term before writing together.
Beyond PromptingRead story
Lead story
A chatbot helped spot false images. The lesson did not stick.
In a study across three countries, a chatbot helped people spot false claims about images right away. A week later, without the tool, the gain was gone.
Source: arXiv · Sep 26, 2026Read original source
Covered by both lenses
Read story
More top stories
Timnit Gebru wants AI scrutiny on harms happening now
In a WIRED interview, Gebru argues that distant AI disaster talk can crowd out questions about data, workers, and misleading sales claims. That is her position, not a settled finding.
AI AgencyRead story
Featured
AI judges need enough shared human checks
When too few answers receive both human and AI review, teams can pick the wrong AI judge. The study tests ways to plan more useful human checks.
AI AgencyRead story
Research & Science
Finding a lead is not the same as proving a discovery
An AI search found a lead near a known enzyme. Whether that counts as a new discovery is in dispute. Lab tests and credit still matter.
AI AgencyRead story
Models & Capabilities
When AI remembers, users need a way to correct it
A paper looks at gaps between what users want an AI to remember and what it keeps. Check what a saved preference covers and how to delete it.
Beyond PromptingRead story
Beyond Prompting
More from this edition
-
Research & ScienceRead story
Parallel AI work needs a real pause button
ParallelPilot studies ways to track several AI coding tasks at once. A visible tracker should let a person pause or redirect work, not just watch it.
-
Culture & Human AgencyRead story
Good human-AI work depends on more than the model
A white paper asks designers to look at the person, AI system, task, workplace, and wider setting. Its robot assembly example shows the method, not proof of better results.