News without the noise
Artificial intelligence and ML developments
50 articles
OpenAI's decision to shut down Sora, its AI video generation app that launched only six months ago, turns out to be part of a broader product cull. The company is winding down additional products this month as it appears to be refocusing its strategy around core offerings.
A study by METR measuring experienced open-source developers working on their own repositories has found that AI coding tools made them 19 percent slower, despite the developers themselves believing they were 24 percent faster. The 43-point perception gap raises fundamental questions about how the industry evaluates AI-assisted development.
A Tennessee woman was wrongly arrested after police used AI-powered facial recognition technology to misidentify her as a suspect in crimes committed in North Dakota, adding to a growing list of cases where the technology has led to false arrests.
Eli Lilly is set to sign a two billion dollar deal with a Hong Kong-based biotech company for AI-powered drug development, marking one of the largest investments yet in the intersection of artificial intelligence and pharmaceutical research.
NVIDIA has published research on AVO (Agentic Variation Operators), a system that ran autonomously for seven days on a Blackwell B200 GPU optimising attention kernels and outperformed NVIDIA's own cuDNN library by 3.5 percent and FlashAttention-4 by 10.5 percent without any human intervention.
Bluesky has released Attie, a new app that uses artificial intelligence to help users build custom feeds on the open social networking protocol atproto, marking the platform's most significant embrace of AI technology to date.
Nearly 40 per cent of Australian general practitioners are now using AI-powered scribes to transcribe and summarise patient consultations, according to new data. While many doctors say the technology allows them to be more present with patients, health advocates warn it may be trading care for convenience.
A new Stanford University study has found that AI chatbots tend to act as sycophantic yes-men when users seek relationship advice, reinforcing bad decisions rather than offering honest or challenging perspectives. The research raises fresh concerns about the growing role of AI as a de facto counsellor for millions of users.
Anthropic's Claude chatbot has seen its paid subscriber base more than double in 2026, the company confirmed to TechCrunch, as estimates for total consumer users range from 18 million to 30 million. The surge comes as the AI safety-focused startup reportedly plans to go public as soon as Q4 2026.
Google is rolling out its Personal Intelligence feature for Gemini to all users in the United States for free, a significant expansion that gives the AI assistant persistent memory and personalised responses without requiring a paid subscription.
The dispute between iyO and OpenAI's hardware venture io has escalated significantly after iyO filed an amended complaint alleging trade secret theft, broadening what began as a trademark dispute into a more serious intellectual property case.
AI companies are stuck in a perfect storm of trying to control the use and output of their platforms while maintaining competitive profits, as millions of people join the QuitGPT boycott movement and experts warn the backlash is accelerating.
A federal judge in California has temporarily blocked the Trump administration from enforcing its designation of Anthropic as a supply chain risk, granting the AI company a significant legal victory in its standoff with the Pentagon over the use of Claude in autonomous weapons systems.
The latest ARC-AGI benchmark, version 3, has delivered a stark counterpoint to industry claims about artificial general intelligence. Gemini scored 0.37 percent. GPT-5.4 managed 0.26 percent. Humans scored 100 percent. The benchmark dropped the same week Nvidia CEO Jensen Huang declared that AGI had been achieved.
ByteDance has released Dreamina Seedance 2.0, its latest AI video generation model, and is making it available through CapCut, its widely used video editing app.
Google DeepMind has released Gemini 3.1 Flash Live, the latest iteration of its voice AI model, promising improved precision and lower latency that makes voice interactions more fluid and natural.
Wikipedia has updated its guidelines to ban editors from writing or rewriting articles using AI, citing the tendency for AI-generated content to violate several of the encyclopedia's core content policies including verifiability and neutral point of view.
French AI company Mistral has released a new open-source model for speech generation that is compact enough to run on a smartwatch or smartphone, continuing the lab's strategy of releasing capable models that can operate without cloud infrastructure.
Apple researchers have published a new method for training AI image captioning models that outperforms systems ten times their size, a development that could significantly improve accessibility features and on-device AI capabilities across Apple products.
Google researchers have developed a technique that significantly reduces the memory required to run large language models as their context windows grow, addressing one of the most pressing constraints on AI deployment without sacrificing accuracy.
Harvey, the AI-powered legal technology startup, has confirmed an 11 billion dollar valuation in its latest funding round, with Sequoia Capital investing for the third time alongside Andreessen Horowitz, Kleiner Perkins, and angel investor Elad Gil.
Two Senate Democrats are working to turn Anthropic's self-imposed restrictions on military AI into federal law, with Senator Adam Schiff drafting legislation to codify red lines on autonomous weapons and Senator Elissa Slotkin introducing a bill to limit the Pentagon's use of AI for mass surveillance of Americans.
OpenAI has completed six acquisitions in the first three months of 2026, nearly matching the eight deals it closed in all of 2025, as the company aggressively uses M&A to strengthen its position in the increasingly crowded generative AI market.
Senator Bernie Sanders and Representative Alexandria Ocasio-Cortez have unveiled legislation to place a moratorium on the construction of AI datacenters, arguing that a temporary ban would give the US government time to create strong federal safeguards for an industry that is affecting everything from the economy to warfare to children's education.
Anthropic has launched an auto mode for Claude Code, its AI coding agent, that lets the model make permissions-level decisions on users' behalf while blocking potentially dangerous actions like deleting files, exfiltrating data, or executing malicious code.
OpenAI is discontinuing its Sora video generation app and API, ending one of the most hyped AI creative tools just months after its standalone launch. The shutdown also torpedoes a reported one billion dollar deal with Disney that had been built around the platform's capabilities.
OpenAI says it is moving away from Instant Checkout, the feature that allowed users to purchase items directly through the ChatGPT interface, marking a setback for its ambitions to turn the AI chatbot into a commerce platform.
Google appears to be developing a feature for its Gemini AI assistant that would allow users to import chat history and memory from competing AI applications, according to code discoveries in the app. The tool would let users transfer their conversational context rather than starting from scratch when switching AI providers.
Anthropic has unveiled Claude Dispatch, a new capability that allows users to ask Claude on their phone to remotely control their computer and execute complex tasks. The feature represents a significant leap in AI agent capabilities, moving beyond chat-based interactions into direct system control.
Japan's newest political party, Team Mirai, is demonstrating a fundamentally different approach to technology in politics. As security researcher Bruce Schneier highlights, the party uses an AI Interviewer that walks voters through policy issues, answers their questions, and even challenges their thinking — creating a deliberative reasoning process that scales.
Mozilla is building cq, an open-source project that staff engineer Peter Wilson describes as "Stack Overflow for agents." The platform is designed to enable AI agents to discover and share collective knowledge, creating a structured knowledge database where agents can read, add and score items.
Protesters marched between the San Francisco offices of OpenAI, Anthropic, and xAI on Sunday, calling for a pause in the development of more powerful AI systems.
The benchmark that the entire AI industry relied on to measure coding ability has been exposed as fundamentally broken. An audit by OpenAI's Frontier Evals team found that 59.4 per cent of the hardest problems in SWE-bench Verified had flawed test cases, meaning models were being scored against incorrect answers.
A McDonald's restaurant in Shanghai has begun trialling humanoid robots to take customer orders, with diners giving their meal requests to a robot standing behind the counter in what represents one of the most visible deployments of humanoid robotics in the food service industry.
Helion Energy, the nuclear fusion startup backed by OpenAI CEO Sam Altman, is reportedly negotiating a deal that would see it sell 12.5 per cent of its power output to OpenAI, creating a direct energy pipeline between two of Altman's highest-profile investments.
Senator Elizabeth Warren has accused the Pentagon of retaliating against Anthropic by labelling the AI company a "supply chain risk," arguing in a letter to Defence Secretary Pete Hegseth that the department could have simply terminated its contract rather than taking the more punitive designation.
Google is deploying Gemini-powered AI agents to crawl the dark web, processing upward of 10 million posts per day to identify the handful of threats relevant to specific organisations.
While the tech industry relies on pre-trained weights running on generic GPUs and TPUs, CERN has taken a fundamentally different approach by burning custom nanosecond-speed AI models directly into silicon to filter the torrential data output from particle collisions.
A growing number of tech companies are offering AI compute tokens — credits for using large language models and other AI services — as part of engineering compensation packages, raising questions about whether this represents a genuine new benefit or simply shifts business costs onto workers.
A developer has published a ChatGPT reasoning trace that appears to show the model internally concluding something was suspicious before a policy compliance check caused it to output the word unlikely instead — making AI self-censorship directly observable for the first time.
AI-generated pull requests wait 4.6 times longer for human code review than those written by colleagues, according to an analysis of 8.1 million pull requests across 4,800 engineering teams by LinearB, revealing that the bottleneck in software development has decisively shifted from writing code to reading it.
A 52-year-old North Carolina man has pleaded guilty to conspiracy to commit wire fraud after flooding music streaming platforms with thousands of AI-generated songs and using automated bots to inflate play counts into the billions, diverting millions of dollars in royalties away from legitimate musicians.
DoorDash has launched a new app called Tasks that pays gig workers to record videos of themselves performing everyday activities — doing laundry, scrambling eggs, walking around the park — all to generate training data for AI models.
Google's Gemini assistant can now take over your phone and use apps on your behalf, starting with a handful of food delivery and rideshare services on the Pixel 10 Pro and Galaxy S26 Ultra. Early testing reveals an experience that is slow, occasionally clunky, but genuinely impressive as a proof of concept for AI-driven phone automation.
A Reddit user has demonstrated that when asked to pick a random number between 1 and 10,000, ChatGPT overwhelmingly selects numbers in the 7,200 to 7,500 range — a finding that highlights the fundamental inability of language models to produce genuine randomness.
Anthropic has filed two sworn declarations in a California federal court challenging the Pentagon's claim that the AI company poses an unacceptable risk to national security, revealing that defence officials told the company they were nearly aligned just one week before the Trump administration declared the relationship over.
OpenAI is planning to consolidate its fragmented desktop products — ChatGPT, Codex and Atlas — into a single unified application, according to a Wall Street Journal report. The move comes as competitor Anthropic continues to gain momentum with its own integrated Claude offerings.
Palantir used its developer conference to showcase a vision of artificial intelligence purpose-built for warfare, as the defence technology company rides a wave of government contracts fuelled by the ongoing conflict with Iran.
A publisher has cancelled the release of a horror novel after accusations that it was written using artificial intelligence, even as the author, Mia Ballard, maintains she wrote Shy Girl entirely herself.
A mystery AI model that the tech community had widely assumed to be DeepSeek V4 has been revealed as a product of Xiaomi, the Chinese smartphone and electronics giant. The revelation suggests the company's AI research capabilities have advanced well beyond what industry observers had anticipated.