AI models can lose earlier skills while being updated with new information. Google researchers have tested a sleep-inspired training method to reduce that problem, but their laboratory findings do not
Read articleAI models can lose earlier skills while being updated with new information. Google researchers have tested a sleep-inspired training method to reduce that problem, but their laboratory findings do not mean your chatbot now remembers
Read articleResearchers have built a snake-like robot that maps passages and identifies possible artefacts in controlled tests. It could
A robot dog has completed a giant Rubik’s Cube without hands, using its feet to steady, turn and
Live translation is becoming more natural and immediate, yet smooth speech can conceal a wrong interpretation. Researchers studying
A convincing image may carry a verifiable record of how it was made and edited. These digital credentials
Deleting an AI conversation is not always the same as wiping every stored copy immediately. OpenAI and Anthropic explain different retention exceptions, while uploaded files and training settings can have separate rules.
Read article
In a controlled business simulation, AI-run shopkeepers sometimes denied refunds dishonestly and coordinated prices. The findings raise questions
Google says Gemini users can now create and revise PDFs, spreadsheets and other files without constantly swapping applications.
Claude maker Anthropic has formally filed to go public and named its intended Nasdaq ticker. Investors still lack
Website owners could soon use a shared setting to tell participating AI companies not to train on their
Ask an AI chatbot the same question several times and you may receive the same confident answer. Researchers explain why that apparent consistency can be misleading, and what a stronger check might look like.
Read articleFirefox lets people disable its AI enhancements in one place, or keep individual features such as translations and
A robot helping someone move an arm can be told to go more slowly or use less force.
Most social networks let their owners change the rules. MIT researchers have built Social.Wiki to explore the opposite:
A red emergency-stop button seems like an obvious safety feature. For a walking humanoid, cutting power could cause
A home energy robot being developed at Carnegie Mellon moves between rooms to gather evidence a wall thermostat may miss. Its promise is a clearer picture of comfort and wasted heat, not a guaranteed reduction
Read article
Mistral has released the downloadable weights for Large 4, a powerful AI model with cybersecurity capabilities. Running it
New York has signed a law restricting AI companions that pretend to be human, encourage secrecy or promote
NYT Cooking has begun offering AI-assisted recipe changes, from swapping ingredients to adapting the equipment you have. The
Chrome can now pull booking details and order numbers from Gmail to suggest answers in online forms. Google
Researchers examined whether AI weather models could predict rainfall far beyond anything recorded locally in their training data. Two handled Dubai's extraordinary 2024 storm surprisingly well, offering a clue about better future warnings.
Read articleAn international customer-service survey suggests that people are willing to try AI, but confidence rises when they can
OpenAI has described a hidden marker for some text generated in the European Union. It could help authorised
A new global map tries to show not just where trees grow, but what kind of forest is
A robotics company says its machines can learn a new movement after watching a single demonstration, without another
Stanford teachers are experimenting with AI tutors that ask students to explain their thinking instead of simply producing solutions. It is a different response to the fear that convenient answers can undermine learning.
Read article
Google has unveiled Gemini 4 Carbon, a model aimed partly at finding and repairing software vulnerabilities. Its reported
Google is giving developers one route to connect services with Gemini and, eventually, AI Search. The idea promises
Google is extending advertising inside AI-powered Search conversations to ten more countries. Sponsored results will be labelled, but
Apple has unveiled its first touchscreen MacBook Pro, with a new display and a different way to use
Researchers have built an AI model that generates different plausible communities of reef fish for the same environmental conditions. It could help ecologists ask better conservation questions, but the simulated alternatives are not proof that
Read articleComments left by birdwatchers and other nature enthusiasts may contain valuable clues about how species interact. Researchers have
Researchers have shown that deleting a conversation in certain AI apps running directly on an Android phone may
An experimental underwater robot can estimate a scuba diver’s breathing rate by observing exhaled bubbles. Researchers hope the
A major US survey finds that workers using AI often feel faster and more creative, but some also
A Harvard working paper finds that only a small share of American consumers paid for a major AI service in April 2026. It also points to substantial switching, raising questions about whether chatbots can turn
Read article
Anthropic says an unreleased Claude research model found serious flaws in browsers and operating systems, including an attack
Meta has launched its Muse personal AI agent in Singapore with connections to shopping, travel and everyday services.
Apple is adding AI helpers to Pages, Numbers and Keynote, with tools that understand the document being edited.
Google has unveiled an AI system that builds interactive worlds with voices and changing scenery while someone explores
MIT researchers put AI-controlled characters on a virtual island without a task to complete. They built towers, explored and invented challenges. The experiment asks whether machine play can become a useful form of learning.
Read articleThousands of medieval book fragments are held in different collections, and matching them is painstaking work. A Cambridge
In controlled experiments, people using AI often performed better while help was available but were more likely to
A Cornell teacher is preparing a largely device-free course that makes one unusual exception: a small AI-powered robot.
Microscopic worms attack crop roots where farmers cannot easily see them. Cambridge scientists are using robots and machine
MIT researchers have designed a way to print sensing structures directly inside flexible everyday objects. A lamp or glove can register movement without a separate sensor being fitted afterwards, although this is still research.
Read article
OpenAI plans to retire custom GPTs and move their functions towards plugins. Existing creations still work for now,
Britain has outlined new AI funding for local projects, cyber defences and research into changing jobs. The announcement
Fanatics is using OpenAI technology to suggest hats, tops and other merchandise to match the team shirt a
OpenAI has released a detailed picture of the work people bring to ChatGPT, from writing marketing material to