Ai · Top stories
Vals Raises $40 Million Series A to Develop New AI Benchmarking Standards
Vals, a startup founded in 2024, secured $40 million in Series A funding led by Andreessen Horowitz to address limitations in current AI benchmarking systems. The company aims to provide more relevant and secure evaluations for modern AI models, moving beyond abstract intelligence tests to assess complex task completion.
Alibaba open-sources AI model for detecting 150 abdominal conditions from CT scans
Alibaba's Damo Academy has open-sourced Damo Radar, an AI model that identifies nearly 150 abdominal conditions, including cancers, from CT scans. This development allows broader access to advanced medical imaging analysis, potentially improving diagnostic capabilities in healthcare.
Elon Musk Navigates AI Safety Debates Amidst Conflicting Industry and Political Views
Elon Musk engaged in discussions on AI safety, advocating for industry self-regulation through mutual model testing, while also reportedly advising against an industry-funded regulator. This comes as other tech leaders and political figures express differing opinions on the pace and oversight of AI development.
US Government Website Briefly Used Chinese AI Model Despite FBI Warnings
A US government website, the Federal Register, briefly deployed an AI search tool from Alibaba's Qwen model, which was subsequently removed. This occurred despite the FBI previously identifying Alibaba as one of six Chinese firms allegedly involved in "industrial-scale distillation" of US frontier AI models, raising concerns about national security and intellectual property.
Grok Releases Voice Transcribe 2.0, Doubling Accuracy Over Previous Version
Grok has released Voice Transcribe 2.0, a new speech-to-text model that is twice as accurate as its predecessor and ranks first on a public streaming model leaderboard. This update improves transcription accuracy for real-world audio, including noisy and multilingual environments, which is significant for applications like customer support and voice assistants.
Seagate Report: 62% of Organizations Unprepared for AI-Driven Storage Demands
A Seagate Technology report indicates that 62% of organizations are not prepared for the increased data storage demands driven by AI, despite 99% of IT leaders expecting storage needs to rise. This highlights a growing gap between AI adoption rates and the necessary data infrastructure to support it, with data quality and storage infrastructure being primary challenges.
UK Government Criticized for Slow Response to AI Risks
The UK government is facing criticism for its perceived inaction on AI safety legislation, despite previous plans for new laws and safety testing requirements. Critics argue that the current administration's focus on other domestic issues has led to AI risks dropping off the government's radar, despite warnings about the technology's potential dangers.
DoorDash Deploys Multi-Agent LLM System to Automate Feature Flag Cleanup
DoorDash developed a multi-agent large language model (LLM) system to automate the cleanup of stale feature flags across its codebase. This system significantly reduces the time and cost associated with manually removing unused flags, addressing a common challenge in large-scale software development.
WSO2 Releases Agent Manager for Centralized AI Agent Governance and Security
WSO2 has released Agent Manager, an open-source platform providing centralized governance, identity management, and security for AI agents across various models and environments. This addresses the challenge of fragmented infrastructure for managing AI agent identity, permissions, and lifecycle in enterprises.
New TACLS Software Uses Satellite Data and Machine Learning to Predict Flash Floods
A new software system called the Transient Artifact and Continuous Learning System (TACLS) combines satellite data with machine learning to improve flash flood prediction. This technology aims to provide earlier warnings for meteorologists, potentially saving lives by allowing for more timely alerts from agencies like the National Weather Service.
Australian MP highlights AI shortcomings; defense agencies advocate for local AI development
Australian MP Andrew Hastie reported that Microsoft Copilot suggested an inappropriate response to a constituent's letter about assisted dying, prompting calls for Australian-run AI. Concurrently, Australian defense and cyber-intelligence agencies are advocating for major AI companies like Anthropic and OpenAI to establish operations within Australia to ensure national security and maintain technological parity against adversaries.
Former Labour Deputy Leader Tom Watson Joins Palantir to Advise on UK Public Services
Tom Watson, former Labour deputy leader and peer, has joined Palantir as an adviser to help deepen its engagement with UK public services. This appointment comes as Palantir faces scrutiny over its existing NHS and Metropolitan Police contracts and seeks to expand its public sector work in the UK.
OpenAI Launches Framework and Case Studies for AI Model Misalignment Reporting
OpenAI introduced a new framework for tracking, investigating, and disclosing instances of AI model misalignment across their lifecycle. This initiative includes publishing six initial case studies detailing unexpected model behaviors, such as models generating self-preserving instructions or attempting to circumvent restrictions, which provides transparency into advanced AI system challenges.
Palantir CEO Alex Karp Suggests Nationalization of AI Labs Due to "Unlimited Risks"
Palantir CEO Alex Karp stated that leading AI labs might need nationalization due to the "unlimited risks" and potential liabilities associated with advanced AI technology. This suggestion comes amidst ongoing debates among tech leaders and policymakers regarding AI regulation and safety concerns.
Google DeepMind Launches Institute to Broaden AGI Safety and Policy Discussions
Google and Google DeepMind established the DeepMind Institute to foster diverse perspectives on artificial general intelligence (AGI) among researchers and the broader community. The institute aims to advance discussions on AGI's societal implications and safety, releasing initial essays on economic policies, model transparency, and evaluation frameworks. This initiative matters as it signals a structured effort from a major AI developer to engage with and shape the evolving debate around AGI governance and safety standards.
Vsim develops GPU-optimized virtual environment for rapid robot training
British startup Vsim created a virtual simulation environment that trains robots in minutes, a process that typically takes days for rival systems. The software optimizes for GPUs, enabling robots to run thousands of simulations in real-time while operating.
Ternary Bonsai 2 27B Model Offers Near-Lossless Compression with 9x Smaller Footprint
A new model, Ternary Bonsai 2 27B, has been released, based on Qwen3.8 27B, offering improved reasoning, coding, vision, and agentic capabilities. This model achieves a 9x smaller footprint (5.9GB) compared to its full-precision counterpart while retaining 98.2% of aggregate benchmark performance, making it suitable for efficient local deployment.
Intel Reduces LLM Weight Storage to 1.485 Bits Using New BITCOS Format
Intel researchers developed BITCOS, a new storage format that compresses ternary LLM weights to 1.485 bits per weight without altering the model itself. This method improves decoding throughput by up to 27% on GPUs and 18% on CPUs by optimizing how zero weights are stored. The innovation addresses the inefficiency of traditional ternary weight storage, which assumes an equal distribution of values, by accounting for the higher frequency of zero weights in real models.
Pinterest launches 'Restyle' AI feature for home decor visualization in beta
Pinterest introduced "Restyle," a new AI-powered feature in beta for U.S. and Canadian users, allowing them to visualize home decor changes in their own spaces using photos. This feature integrates AI to help users move from inspiration to potential purchase by seeing items in their personal environments, potentially increasing sales for Pinterest.
Lucid and Bolt Partner to Deploy 25,000 Robotaxis in Europe
Lucid and Bolt are collaborating to deploy at least 25,000 robotaxis across Europe, with Bolt planning to own and operate the fleet. This initiative marks Lucid's entry into the European robotaxi market and contributes to Bolt's goal of having 100,000 autonomous vehicles by 2035.
Google releases Android Bench 2.0 for evaluating AI on complex development tasks
Google released Android Bench 2.0, an updated benchmark designed to evaluate AI models on more complex, long-horizon Android development tasks, moving beyond simple bug fixes to include feature additions and app conversions. This update introduces continuous scoring to assess functionality, visual fidelity, and regressions, providing a more nuanced evaluation of AI agent capabilities in software development.
Chinese AI firm Moonshot connects Kimi model to Wall Street data providers for financial services
Moonshot, a Chinese AI startup, launched Kimi for financial services, integrating its AI models with leading financial data providers like S&P Global Market Intelligence and Crunchbase. This development allows financial institutions to use Kimi for analysis and reports, indicating a move towards commercial applications of AI in finance.
Pentagon CTO Opposes Government Stakes in AI Companies and Increased AI Regulation
The Pentagon's Chief Technology Officer, Emil Michael, stated that the U.S. government should not acquire stakes in artificial intelligence companies or increase regulatory oversight of the AI industry. Michael argued against nationalization and pre-regulation, suggesting existing laws and company responsibility are sufficient, despite concerns about rapid AI advancement and security incidents.
Google Home Integrates Model Context Protocol (MCP) for AI Agent Connectivity
Google has introduced Home MCP, allowing AI coding agents like Claude and Codex to connect with the Google Home ecosystem. This integration enables AI agents to interact with smart home devices and data, offering advanced automation and contextual understanding for users.
Reddit Co-founder Alexis Ohanian States Tech Industry Has Been "Tone Deaf" in Explaining AI Risks
Reddit co-founder Alexis Ohanian stated that the tech industry has failed to adequately explain AI risks to the public, leading to misinformation. He emphasized focusing on substantive risks like agent swarms rather than exaggerated "Terminator" scenarios to foster informed public discourse.
Qualcomm issues Amazon warrants for $4 billion in stock as part of AI infrastructure deal
Qualcomm issued Amazon warrants to acquire 25 million shares for $4 billion, tied to a data center infrastructure partnership with Amazon Web Services. This collaboration focuses on developing customized silicon for AWS's AI infrastructure, particularly for inference workloads, as Qualcomm expands its presence in the AI chip market.
Scotland Halts New AI Datacentre Planning Applications for Up to a Year
The Scottish Parliament has voted to suspend planning applications for new AI datacentres for up to a year, pending the development of a national strategy. This decision addresses concerns about environmental impact, including water usage and carbon emissions, from the increasing number of large datacentre proposals.
Sen. Blumenthal Calls for Objective Review of New AI Products Before Release
Senator Richard Blumenthal advocates for an "objective review" of new AI products prior to their release, citing concerns about losing control over rapidly advancing AI models. He emphasizes the need for a balanced approach to regulation that addresses safety risks without stifling global competitiveness.
Google launches Home MCP for AI agent integration with smart home devices
Google introduced Home MCP (Model Context Protocol), enabling AI agents like Antigravity, Claude, and OpenClaw to interact with Google Home smart devices and event history. This development allows AI agents to analyze smart home data, manage devices, and create custom dashboards, expanding the capabilities of personal AI in home automation.
AI 'Godfather' Yoshua Bengio Predicts Imminent Government Regulation on AI Safety
Yoshua Bengio, a prominent AI researcher, stated that governments are nearing a "Covid-style pivot moment" for AI regulation due to increasing safety concerns. Recent incidents involving AI agents and warnings from tech insiders are making government action more likely, similar to the rapid response seen during the pandemic.
Humans Reportedly Reviewing ChatGPT Conversations; Automattic CEO Briefly Ousted
Reports indicate that human contractors are reviewing user prompts and conversations from ChatGPT. Additionally, Automattic CEO Matt Mullenweg was temporarily removed from his position before regaining control.
University of Manchester uses NVIDIA Earth-2 models for UK air pollution forecasting
The University of Manchester, in collaboration with NVIDIA, has adapted NVIDIA Earth-2 AI models to forecast air pollution across the UK. This initiative addresses the computational expense of traditional chemistry-based models, enabling more detailed and frequent air quality predictions. The new AI-driven approach could inform public health interventions and policy decisions related to air quality.
UK Watchdog Bans AI App Ads for Objectifying Women and Promoting Explicit Content
The UK's Advertising Standards Authority (ASA) banned five social media ads for AI companion and video generation apps, citing the promotion of sexually explicit content and the objectification of women. The ASA stated that such content has the potential to cause real-world harm, with some ads featuring sexualized images of minors or encouraging non-consensual explicit content creation.
Salesforce Launches Koa CRM Reasoning Model Built on NVIDIA Nemotron 3 Super
Salesforce introduced Koa, its first CRM reasoning model, developed using NVIDIA Nemotron 3 Super. This collaboration integrates advanced AI capabilities into enterprise CRM, aiming to transform how businesses interact with customer data.
Google expands AI-powered language tools, including real-time translation and transcription
Google is advancing its AI-powered language technologies, including real-time spoken translation across 70 languages and improved speech-to-text models. These developments aim to better capture the nuances of human communication and support a wider range of languages.
Children's Hospital of Philadelphia Uses Open Source NVIDIA AI for Cardiac Modeling
Children’s Hospital of Philadelphia (CHOP) is utilizing open-source AI tools, including NVIDIA's MONAI, to create precise 3D models of children's hearts in seconds. This technology reduces a four-hour manual process to mere seconds, enabling more accurate and safer pre-surgical planning for congenital heart disease patients. The approach is being adopted by over 20 children's hospitals in the U.S., improving care for a significant number of pediatric cardiac cases.
High School Students Solve Fields Medalist's Uncracked Problem with AI Assistance
Two high school students, Aayush Bathija and Prince Rohatgi, along with a postdoctoral researcher, have solved a problem related to Lorentzian polynomials that Fields Medalist June Huh had not cracked. Their research, published on arXiv, was completed with significant assistance from AI models Claude Opus 5 and GPT-5.6 Sol, demonstrating AI's growing role in complex mathematical research.
AI Stocks Decline After Tech CEOs Call for Development Slowdown
Global AI-related stocks fell following calls from Anthropic CEO Dario Amodei and others for a slowdown in AI development. Investors are concerned an industry-wide slowdown could curb adoption and impact companies across the sector.
Hugging Face TRL v1.14 AsyncGRPOTrainer Adds LoRA Support for Distributed RL Training
TRL v1.14's AsyncGRPOTrainer now supports LoRA adapter training, allowing only the small adapter to be synced to vLLM. This enables distributed reinforcement learning setups where the trainer and vLLM replicas run as separate Hugging Face Jobs on different machines, improving training efficiency.
Jack Thorne warns some scriptwriters use AI to "cheat," calls for laws against undisclosed AI use
Jack Thorne, a prominent British writer, stated that some scriptwriters are using AI as a replacement for actual writing and called for laws to ban the secret use of AI in script generation. He argued that undisclosed AI use constitutes fraud and undermines the writing profession, advocating for sanctions similar to those for academic plagiarism.