AInauten News - curated AI news

Evaluating Microsoft Copilot at enterprise scale requires test data that captures the full complexity of real organizations, yet real enterprise data contains sensitive information that cannot be used for evaluation under any circumstances. The post From synthetic enterprise data evaluations to real Microsoft Copilot AI confidence appeared first on Microsoft Research.

Research finds 55% of lawyers are using AI in legal practices, up from 37% last year, though efficiency gains aren’t as high as expected Get our breaking news email, free app or daily news podcast More than half of Victoria’s lawyers have reported using AI in their work, according to a new study, but just over a quarter say they’ve reduced fees as a result of efficiency gains. They also say clients seeking advice from chatbots has added to their workloa…

Artificial intelligence has undergone a profound shift in how it learns, moving from rigid teacher-driven methods to more flexible, student-driven approaches. This evolution has been particularly impactful for training weaker models, which often struggle with traditional methods like supervised fine-tuning.

Host Sowbhagya Suresh shares some of the latest news and cutting-edge research from across Microsoft—from understanding the challenges facing coding agents to uncovering how people use AI at work and improving extended-range weather forecasting. The post Microsoft Research Focus – Edition #97 appeared first on Microsoft Research.

Three-quarters of all responses were machine-generated in failed attempt to stop bill from passing into law, research finds AI bots flooded a government consultation with tens of thousands of fake responses in a failed attempt to stop the UK’s ban on disposable vapes. Three-quarters of all the responses to the 2023 consultation on the tobacco and vapes bill were machine-generated, in what is thought to be the first coordinated attempt to use AI to derai…

Harness², developed collaboratively by Google Research and Northwestern University, introduces a new approach to AI adaptability by focusing on self-improvement without additional training. Unlike static systems that can falter in specialized tasks, Harness² uses a self-evolving mechanism to refine workflows dynamically.

Anthropic has filed to go public on the US stock market, publishing audited financials for the first time. The S-1 prospectus, released on Monday, shows explosive but extremely expensive growth. More than 90 pages are devoted to risks that could hurt the business, ranging from cyber-attacks to the limits of the company's own safeguards.

In this episode of Merryn Talks Money, Merryn Somerset Webb and Steve Hou, head of research at Silicon Data, look at an emerging problem in AI economics: token prices are falling while the cost of the compute needed to produce them is rising. They discuss what this mismatch could mean for the valuations of companies like OpenAI and Anthropic, whether demand can justify the enormous sums invested in AI infrastructure, and what happens when everyone has a…

Artificial intelligence has been embedded across the UK Home Office in near-total secrecy, reinforcing the so-called hostile environment for immigration, according to a report by Open Rights Group (ORG). Tasks from detecting potential sham marriages to determining the age of asylum seekers, border surveillance and checking the contents of phones belonging to new arrivals now involve the technology, the report says.

Multi-Turn Reinforcement Learning (MTRL) on Amazon SageMaker AI trains models on the outcome of a whole multi-step task rather than labeled steps. This hands-on guide walks the full pipeline using Wordle: deploying the agent and reward on Amazon Bedrock AgentCore, preparing the dataset, training, and evaluating at serving settings.

WA paid Deloitte $166,000 for flawed feasibility study on merging the University of Western Australia, Curtin university and Murdoch university Get our breaking news email, free app or daily news podcast The Western Australian government has said it is “extremely disappointed” and is seeking an “expedited” cost recovery from Deloitte after errors were discovered in a report on a potential university merger, including AI-hallucinated references and hundr…

The next competitive advantage in enterprise AI depends less on which model a company chooses than on what it builds around it, MIT Technology Review argues. For several years the AI conversation centered on model capability: bigger models, better reasoning, larger context windows and higher benchmark scores.

Microsoft's new AI diagnostic system beat generalist doctors on hard cases, according to a study published in Science, TechCrunch reports. Microsoft AI chief Mustafa Suleyman says the system outperforms doctors on difficult diagnoses. But most of the improvement appears to come from the underlying AI models, with Microsoft's own orchestration system adding comparatively little on top.

Microsoft claims it has built an AI-powered tool that can diagnose conditions up to four times more accurately than human doctors, The Verge reports. The findings, published in Science, are based on tests pitting Microsoft's AI system against physicians on scenarios drawn from medical case reports and emergency department data. Microsoft compiled the cases into its own benchmark.

Microsoft Research has introduced Flint-interactive, a language that lets AI agents create interactive data visualizations. Users can click a chart, filter data, highlight related content and explore beyond the view the author originally chose. The post presents the research project and its capabilities but offers little information on availability, integration effort or performance with large datasets.

Bloomberg Technology's The Close covered the latest Wall Street session as Tesla, IBM and Meta reported quarterly earnings with AI spending at the center of investor attention. Guests included BlackRock's Armando Senra, CME Group Chief Executive Terry Duffy, Bose Chief Executive Lila Snyder, Wedbush analyst Ygal Arounian, ARK's Tasha Keeney and former Tesla board member Steve Westly.

Anthropic has announced Claude Duet, an AI agent for professionals and small business owners that competes with Meta's Muse and OpenAI's Dots. Duet launches as a research preview, available as a standalone app on iOS and Android and through Claude on the web, and is built on Claude Opus 5.5. Each morning it sends a short brief on the most important items across connected apps.

A team on Hacker News has launched a search engine built for AI agents that attaches a usefulness score and a price to every result page. The idea came from testing agents that research companies for sales outreach: in four out of seven cases they produced outdated insights even when recent news was in their context, and a larger model did not fix it.

It is now possible to build AI chatbots of loved ones from their texts, voicemails and other personal data and keep talking to them after their death. A global research study is examining how these deadbots, or postmortem avatars, affect people who are grieving, the legacy and rights of the deceased, and grief practices in different cultures. Vox explores the questions in a new episode of its Unexplainable podcast.

In a Guardian opinion piece, public health expert Devi Sridhar questions whether AI can predict cancer survival better than doctors. Researchers at Mass General Brigham have developed an AI model that estimates a cancer patient's prognosis from a single photo of their face. Sridhar calls the research promising and notes that clinicians have long used facial appearance as a rough proxy for health.

Doctors at Mass General Brigham, a nonprofit academic healthcare system in Massachusetts, have developed an AI tool that predicts a cancer patient's chances of survival from a single photograph of their face, The Guardian reports. The custom-built system estimates a person's biological age from the image and uses it to directly predict survival outcomes for patients diagnosed with cancer.

The average person spends about 40% of their waking hours online, producing the equivalent of 229kg of carbon emissions a year, according to a study in Nature Communications cited by The Guardian. The piece argues that photo and video archives, overflowing inboxes and unread subscriptions all add to demand for ever larger digital infrastructure, at a time when many Australians worry about datacentres consuming land and energy.

An accelerating sell-off in AI-linked shares has hit global markets, The Guardian reports. Investors are worried about rising inflation and possible interest rate hikes, which weigh especially on a heavily indebted sector. Chip makers and tech investors such as Samsung and SoftBank fell sharply.

Pledge spanning five years from world’s most valuable firm comes as concerns about AI agents have risen Business live – latest updates The $5.5tn (£4.1tn) US chipmaker Nvidia is donating $250m worth of computing power to the UK to support AI safety research. The donation from the world’s most valuable company comes as the UK government confirmed its AI Security Institute (AISI), which tests advance AI models, will be able to examine cutting-edge technol…