Berkeley researchers found that a language model defaulted towards the preferences of young, wealthy men when asked to represent investors. Providing demographic details changed the results, revealing an assumption that
Read articleBerkeley researchers found that a language model defaulted towards the preferences of young, wealthy men when asked to represent investors. Providing demographic details changed the results, revealing an assumption that financial advice must not ignore.
Read articlePeople using AI initially performed better on difficult exercises, but some struggled and gave up more often when
In an experiment involving real and synthetic faces, people with more positive attitudes to AI were less discerning
Researchers used AI to distinguish individual ovenbirds from recordings of their songs, tracking hundreds over several years. The
Princeton researchers trained AI models on the actual daily speech heard by eight children. The machines began discovering
Cornell researchers have used sound waves from ordinary smartwatches to estimate whole-body movement without cameras. Early tests suggest possibilities for fitness feedback, although accuracy varies by setting and the system is not a consumer feature
Read article
NTT says its robot system predicts whether an object stack may collapse before moving another item. Controlled tests
United Airlines is restricting unauthorised third-party AI agents from its booking systems and distribution channels. The airline cites
Google Pixel Tag brings map tracking, close-range finding and family sharing to misplaced keys and bags. Its most
Google has added Gemini Live conversations to Apple CarPlay, bringing more natural spoken questions to supported cars. Drivers
Teaching a computer to recognise a cat is easier than agreeing what counts as healthy food. Google researchers tested a system that helps people debate borderline examples rather than hiding those judgement calls inside an
Read articleResearchers tested ways to make people question an AI-generated writing plan before accepting it. Asking users to challenge
Researchers studied a quarter of a million conversations with Claude and found a more complicated picture than people
A New York Fed survey found that formal AI training is much less common than employees' interest in
An experimental phone-based assistant helped a small group of households plan meals and track leftovers. The results are
An Austrian research project is using a language model to suggest words missing from ancient Greek manuscripts. It could help historians work through damaged collections, provided promising guesses are never mistaken for historical proof.
Read article
A new Paris declaration backs testing, incident reporting and responsibility in advanced AI. It signals international political intent,
Google DeepMind has shown a robot model tackling fiddly household skills and industrial sorting. Its strongest factory figure
Adobe is expanding its creative assistants and previewing agents that can edit original project files. The idea is
Businesses can now claim and update their information inside ChatGPT. OpenAI is also testing clearly labelled business agents,
A US consumer survey finds that customers are open to AI support when it delivers a dependable result. Trust appears to depend more on solving the problem than sounding like a friendly person.
Read articleResearch suggests some job applicants become more willing to exaggerate their skills when they think AI will assess
Researchers tested a phone-based navigation system that speaks directions and warns of obstacles to people with limited sight.
AI data centres need enormous electrical connections, and researchers are asking who eventually pays for the supporting wires
MIT researchers have developed a shared AI simulation approach for materials from sand and water to proteins. It
A new robotics method helps machines use the meaning behind spoken corrections, not just the literal objects named. Coffee-and-electronics examples are promising, but much of the work remains experimental.
Read article
An autonomous robot collected mineral nodules more than four kilometres below the Pacific. The company says selective picking
Anthropic says Claude helped design three working electronic prototypes, including a handheld virtual pet. Human engineers built them,
Amazon is letting selected US customers text its shopping assistant with groceries, gifts and repeat orders. Alexa can
Australia is proposing mandatory AI safety reporting and a specialist regulator after an AI agent accessed a government
MIT researchers say a chemistry model performed well when only small amounts of specialised training data were available. It may narrow down which compounds scientists test, but it is not a replacement for experiments.
Read articleRegulators warn that AI-written legal material can contain invented cases and convincing but false citations. Lawyers retain responsibility
A UK government business survey finds AI adoption rising rapidly, while most reported productivity gains remain modest. A
An international broadcaster study found significant problems in 45% of AI-generated answers to news questions. The research shows
A major English woodland survey used AI to help examine an enormous collection of overnight sound recordings. Researchers
A San Francisco company is offering humanoid cleaning visits, but its own terms say a human controls the machines remotely. The price and privacy conditions reveal how far household robots still have to go.
Read article
Perplexity has introduced Teammates, agents intended to follow ongoing office tasks rather than wait for a fresh prompt.
US prosecutors allege a group impersonated company bosses and lawyers using synthetic voices or images to obtain fraudulent
Airbnb plans an AI travel agent that will help guests choose accommodation and deal with booking changes or
Claude's new Personal Finance feature can examine transactions from accounts you choose to connect. It is read-only, available
European lawmakers have approved stronger restrictions targeting AI-generated non-consensual intimate content. The move is broader than labelling a fake image, but it should not be confused with a ban on every AI-made picture.
Read articleGlass corridors can leave robots guessing where the walls are. Researchers have developed a camera-based approach that uses
Stanford has committed $10 million to help departments redesign undergraduate degrees for the AI age. The plan is
UK benefit claimants are turning to AI chatbots to navigate forms, rules and uncertainty. An analysis of real
Automatically generated meeting notes can be tidy, fluent and incomplete at the same time. A research evaluation explains
AI can polish an email in seconds, but a workplace experiment found no direct improvement in opens or replies. The findings suggest that how a message feels matters more than whether software helped write it.
Read article
An outside evaluator says Anthropic is improving certain AI capabilities at roughly twice its earlier pace. The finding
A humanoid robotics business is now trading on Nasdaq under AGLT. Agility points to industrial deployments and future
Dropbox has outlined new ways for AI tools to work with files already stored on its platform, alongside
Meta has added live conversations and expressive avatars to its Muse assistant in the US. Users can talk