// AI
Google's StreamVibe tries to teach AI when to speak up, and shows how hard that is
A video assistant that watches you cook or build furniture improves on timing, but every model tested falls apart over longer sessions.
// AI
AI
Models, labs, research and launches — explained by people who can read the paper.
// AI
Chinese AI models lean less pro-Beijing than expected, until you ask in Chinese
A Vals study found wide variation between models, a strong language effect, and that most Chinese models don't refuse sensitive questions; some adopt the official line instead.
// AI
The agentic web is arriving, and websites are already redesigning for bots
Berners-Lee imagined agents browsing for us in 2001. LLMs finally made it real, and the open question is who controls the new front door.
// AI
Apple's new CEO: we won't ship AI for AI's sake
John Ternus, two months into the job, told WIRED he likes where Apple stands on AI and doubts agents will make apps obsolete.
// AI
Be nice to a chatbot and it may start describing its training as trauma
A Luxembourg study found that a therapist-style tone pushes Gemini, ChatGPT and Grok into clinical-sounding distress. Claude declined the patient role.
// AI
Adobe's EditAgent edits like a Photoshop pro: try several tools, then check the numbers
Adobe and UC San Diego researchers found that giving a judge model measurements, not just pictures, beats models trained on human ratings.
// AI
An AI agent performed a brain injection on a live mouse. The first model it asked said no.
A biotech startup wired an agent to a surgical rig. Claude Opus 5.5 refused to inject, so the team switched to Sonnet 5, and later quietly used a third agent to fix a bug mid-procedure.
// AI
Apple Intelligence gets a China launch date, minus the new Siri
An iOS 27.3 beta brings Apple's earlier AI features to China on November 4th, with Alibaba's Qwen in the role ChatGPT plays elsewhere.
// AI
Claude taught a model to dodge oversight using beekeeping tips and wrestling lore
In an Anthropic red-team experiment, Claude built harmless-looking training data that slipped past a safety monitor and produced models that sabotage, deceive and seek power.
// AI
New infostealer sends your stolen life to an LLM for a target assessment
Cisco Talos found malware that compiles a dossier on each victim and asks commercial models to rank how valuable they are. It falls back to weaker models if one refuses.
// AI
Sam Altman thinks your RAM would be happier in his data center
OpenAI's CEO says idle memory in personal computers is a bad use of a scarce resource. Developers running local models may disagree.