Ai · Top stories
Google Antigravity introduces /boost for deep reasoning in software engineering tasks
Google Antigravity launched /boost, a multi-agent reasoning pipeline activated by a slash command, to address complex software engineering problems. This tool breaks down difficult tasks, delegates work to specialized subagents, and verifies solutions iteratively, aiming to improve developer productivity for high-intensity coding sessions.
Pro-AI Super PAC Funds Ads to Promote Data Centers in Key Midterm States
Build American AI, a new group affiliated with the super PAC Leading the Future (LTF), plans to launch a multi-million dollar ad campaign in Kansas, Ohio, and Wisconsin to advocate for data centers. LTF is heavily funded by Marc Andreessen, Ben Horowitz, and OpenAI President Greg Brockman, and supports candidates favorable to AI while opposing those against it. This initiative highlights increasing political spending by tech investors to influence public perception and policy regarding AI infrastructure.
Google Releases TimesFM-3, a New Time-Series Forecasting Model, Under Non-Commercial License
Google launched TimesFM-3, a 330-million-parameter time-series forecasting model trained on over a trillion data points, now available on Hugging Face with a non-commercial license. This model is Google's first natively pre-trained for multivariate time series and zero-shot generalization, outperforming previous models in benchmarks. Its non-commercial license restricts immediate business use, but it advances the field of time-series forecasting.
AI Tool Detects Heart Disease from ECGs in Under Two Seconds
Doctors have developed an AI tool that can identify signs of heart failure and heart valve disease from standard electrocardiogram (ECG) results in under two seconds. This development could accelerate early diagnosis for these common heart conditions, potentially reducing patient wait times for further diagnostic tests like echocardiograms.
AI Hallucinations Infiltrate Australian Parliament Inquiry Submissions
Australia's parliamentary inquiry process is receiving AI-generated submissions that invent studies and misattribute research to real academics. This phenomenon risks parliamentarians making decisions based on fabricated evidence, exacerbated by AI search summaries that validate these fake references.
NHS watchdog warns AI scribes misdiagnose and incorrectly record drug names
Healthwatch England has issued a warning that AI scribes used in NHS consultations are making errors in transcribing drug names and diagnoses, posing a risk to patient safety. These inaccuracies, if undetected, could enter medical records and negatively affect patient care, despite the NHS's rapid rollout of the technology.
Ibteda Digital Library digitizes 1,800 Urdu books using budget cameras and custom AI
A team of Pakistani archivists, operating as the Ibteda Digital Library, digitized 1,800 rare and out-of-print Urdu books over a decade using budget Nikon cameras, accumulating 902,000 shutter clicks. They developed a machine learning process to automate post-processing of 526,000 scans, addressing challenges posed by Urdu script and varied book conditions.
Australian Fair Work Commission Condemns AI-Generated Legal Advice in Dismissal Case
The Australian Fair Work Commission (FWC) has criticized an ex-ALDI worker for using "plain wrong" AI-generated legal advice in a failed dismissal challenge, ordering him to pay legal costs. This incident highlights a 40% increase in FWC cases, partly attributed to litigants using AI, prompting new disclosure requirements for AI use from October 20.
UK Risks Lagging in AI Due to Slow Telecoms Upgrades, Executives Warn
Senior industry executives warn that the UK's slow progress in upgrading its telecoms infrastructure, particularly 5G mobile networks, could hinder its competitiveness in the global AI race. The current network performance is already suboptimal and will struggle to handle the increased traffic from widespread AI adoption.
Tencent Releases and Open-Sources Hy4 Preview Large Language Model
Tencent has released and open-sourced Hy4 preview, a new large language model with 770B total parameters and a context window exceeding 1M tokens. This model is designed for real-world productivity tasks across coding, office work, and scientific research, positioning it among top-tier open-source models.
Salesforce CEO Dismisses "SaaSpocalypse" Fears, Citing AI Integration and Strong Earnings
Salesforce CEO Marc Benioff stated that fears of a "SaaSpocalypse" due to AI are unfounded, following a strong earnings report and better-than-expected guidance. Benioff argues that AI strengthens rather than cannibalizes SaaS businesses, as evidenced by increased spending from leading AI companies on Salesforce platforms and a new partnership with Anthropic.
Labour Rejects Green Party's Call to Halt AI Datacentre Construction in UK
The Labour party has rejected a proposal from Green party leader Zack Polanski to pause the construction of new AI datacentres in the UK. This rejection comes amidst concerns over energy and water consumption, with datacentre electricity use projected to quadruple by 2030.
Open ASR Leaderboard Adds Hindi and Indian English to Address Bias in Speech Recognition
The Open ASR Leaderboard has integrated two new evaluation sets, Monsoon en-IN and Monsoon hi-IN, to include Hindi and Indian English, marking the first Global South languages on the platform. This addition aims to address known biases in automatic speech recognition (ASR) systems, which often perform poorly for non-European languages and diverse speaker demographics.
Netflix implements multimodal embeddings for improved asset personalization at scale
Netflix has integrated multimodal embeddings into its personalization models to address the cold-start problem for new titles and assets. This allows models to understand visual and audio cues of assets, enabling personalized recommendations sooner after a title's launch by leveraging taste signals from related content.
Meta India and Southeast Asia VP Sandhya Devanathan joins OpenAI
Sandhya Devanathan, Meta's Vice President for India and Southeast Asia, has left Meta to join OpenAI. She will oversee consumer growth, enterprise adoption, partnerships, regulatory engagement, and operations across Southeast Asia and Australia for OpenAI, reporting to Kiran Mani.
AI identifies heart disease in women using routine mammograms
Researchers presented findings at the European Society of Cardiology’s annual congress showing that AI can detect coronary heart disease, high blood pressure, and stroke risk in women from standard mammograms. This development could enable dual-purpose breast screenings, helping to identify cardiovascular issues in women who often have undiagnosed heart disease.
DeepSeek Seeks External Funding as Founder's Quant Fund Navigates IPO Volatility
DeepSeek, a Chinese AI lab, is seeking outside investment for its growth, moving beyond its initial financing from founder Liang Wenfeng's High-Flyer Quant hedge fund. This shift occurs as High-Flyer faces revenue instability and increased volatility in the Chinese IPO market, despite securing pre-IPO placements in key tech firms like CXMT and Unitree Robotics.
Google DeepMind introduces double-blind evaluation for AI models using confidential computing
Google DeepMind has developed a double-blind evaluation method for AI models, allowing testing without either the model provider or the evaluator seeing the other's proprietary data. This method addresses benchmark leakage concerns by using confidential computing to protect both model weights and test questions during evaluation. The approach aims to provide a more secure and unbiased way to assess AI model performance.
AI-generated 'ghost' authors contaminate academic publishing and online content
A new preprint research paper from Samsung and the University of Warsaw reveals that AI models consistently generate specific non-existent names, such as Elena Vasquez and Marcus Chen, which are then appearing as authors on hundreds of academic papers and online documents. This phenomenon highlights a growing issue of AI-generated content infiltrating and potentially compromising the integrity of academic and informational sources.
Gemini Omni 1.1 Flash Adds Creative Controls and Generative Video Capabilities for Developers
Google has updated Gemini Omni 1.1 Flash with new creative controls and generative video capabilities, making it production-ready for professional use via the Gemini API in Google AI Studio. These updates allow developers to extend video scenes, specify first and last frames for transitions, and draft videos more efficiently at lower resolutions and costs, enhancing control and iteration speed for generative video workflows.
NVIDIA MPS and Triton on Amazon EC2 Reduce ASR Inference Costs by 75%
AWS, NVIDIA, and Heidi Health demonstrated a 75% reduction in ASR inference costs on Amazon EC2 by using NVIDIA CUDA Multi-Process Service (MPS) with NVIDIA Triton Inference Server. This setup allows for more efficient GPU utilization, reducing the number of required GPU instances from 16 to 4 for Heidi Health's clinical consultation processing. The improved efficiency addresses the challenge of low GPU utilization during ASR inference while maintaining sub-second latency.
Simular's Sai Agent Achieves 73% Success on OSWorld 2.0 Benchmark
Simular's Sai agent scored 73% on the OSWorld 2.0 benchmark, outperforming OpenAI's GPT-5.6 Sol and Anthropic's Opus 5. This result indicates progress in AI agents handling complex, real-world professional tasks at a lower operational cost.
Cocomelon Studio Moonbug Entertainment Directs Animators to Experiment with AI
Moonbug Entertainment, the studio behind Cocomelon, has instructed its animators to begin experimenting with AI tools for content creation. This move indicates a major children's entertainment company is exploring AI integration, albeit with guardrails due to legal and copyright concerns.
Waymo Executive States Cameras Alone Are Insufficient for Full Autonomous Driving
Waymo's VP of Onboard Software, Srikanth Thirumalai, stated that cameras alone are not enough for safe, fully autonomous operations, advocating for a multi-sensor approach including lidar and radar. This statement indirectly challenges Tesla's camera-only strategy for its autonomous driving systems. The differing approaches highlight a fundamental debate within the autonomous vehicle industry regarding sensor reliance.
MIT Committee Releases Report on AI Use in Teaching, Learning, and Research Training
MIT's Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training has released a report outlining its findings and recommendations for integrating generative AI into the institution's educational mission. The report emphasizes the need for MIT to address deeper questions about the structure and value of education in the AI era, providing principles and actions for instructors and administration.
Google Introduces Gemini 3.5 Transcribe, Enhancing Voice Input Capabilities
Google has launched Gemini 3.5 Transcribe, a new AI model designed for faster and more accurate voice-to-text conversion, directly competing with OpenAI's GPT-Live. This development signifies a renewed focus on voice input as a primary interaction method across various Google platforms and for third-party developers.
LAION Releases LAION-BVD, a 1.3 Billion URL Open Video Dataset for Multimodal AI Training
LAION has released LAION-BVD, a large-scale open video dataset containing 1.3 billion platform-specific video URLs, from which 80 million videos totaling 10 million hours were downloaded. This dataset is designed to support multimodal pre-training across video, audio, and image modalities, providing an open resource for academic research in AI.
London neurosurgeons perform first AI-assisted brain tumor removal, saving patient's sight
Neurosurgeons in London successfully performed the world's first AI-assisted operation to remove a brain tumor, using the system to identify critical structures during surgery. This marks a new application of AI in delicate surgical procedures, potentially improving precision and patient outcomes in neurosurgery.
Perplexity introduces Portable Computer, separating AI reasoning from deterministic execution
Perplexity released Portable Computer, a local-first version of its Computer agent, which separates probabilistic AI reasoning from deterministic software execution. This architectural change allows for improved reliability and performance in AI agents by using code for control and policy enforcement, rather than relying solely on larger AI models for orchestration.
Google Cloud's Knowledge Catalog Integrates Open Knowledge Format (OKF) for Agent Context
Google Cloud's Knowledge Catalog now integrates with the Open Knowledge Format (OKF), allowing organizations to share and govern OKF bundles across their systems. This integration enables discoverability, governance, and secure access for AI agents to contextual information, addressing the scalability challenges of managing OKF bundles independently.
Particle launches Radar, a podcast search engine for AI agents and data analysis
Particle, an AI newsreader startup, has launched Radar, a podcast search engine that transcribes and semantically understands podcast audio. Radar provides an API for AI agents and data resellers, with hedge funds being early adopters for accessing unstructured audio data.
WebMCP Standard Proposed for AI Agent-Website Interaction
Google and Microsoft are developing WebMCP, a proposed web standard that allows websites to declare their functionalities as structured tools for AI agents. This initiative aims to replace current screen-scraping methods with a more reliable and efficient way for AI to interact with web pages, improving automation and reducing errors.
Perceptron, founded by ex-Meta scientists, launches Isaac 0.5 for industrial visual AI
Perceptron, a startup founded by former Meta research scientists, launched Isaac 0.5, an open-weight visual AI model designed to help robots perceive, reason, and act in industrial environments like warehouses and factory floors. This release matters as it aims to provide general-purpose AI for physical robotics, addressing limitations of existing specialized or resource-intensive models.
Diagrid Catalyst 2.0 Introduces Durable and Verifiable Execution for AI Agents
Diagrid released Catalyst 2.0, adding failure recovery and cryptographic verification capabilities to AI agents built with various frameworks. This update allows AI agents to resume interrupted multi-step operations without re-executing completed work and provides verifiable proof of execution history, addressing issues of reliability and accountability in AI agent operations.
Bitcoin Mining Firms Pivot to AI Due to Falling Crypto Rewards and Infrastructure Alignment
Bitcoin mining companies are reallocating their computing power and data center infrastructure to artificial intelligence operations, driven by a decline in Bitcoin's value and mining rewards. This shift leverages their existing expertise in managing large-scale data centers and securing cheap electricity, positioning them to capitalize on the growing demand for AI compute resources.
China bans AI companions for minors and restricts adult chatbots due to social concerns
China has implemented new regulations banning AI companions for minors and placing restrictions on chatbots marketed to adults, citing concerns about emotional dependence and the replacement of social interaction. This move follows the shutdown of ByteDance's Doubao companion feature and highlights growing government apprehension regarding AI's social impact amidst its widespread adoption in China.
Perplexity's Computer AI agent now runs locally on high-end Nvidia hardware
Perplexity, in collaboration with Nvidia, released Portable Computer, a local version of its agentic AI assistant. This development allows users to run the AI agent on their own hardware, requiring specific Nvidia GPUs with at least 24GB VRAM or a DGX Spark desktop, which entails significant hardware costs.
DuckDuckGo Survey Reveals Users Share Significant Personal Data with AI Chatbots
A DuckDuckGo survey found that nearly one in three AI chatbot users share information they would not tell others, with 53% unaware that most chatbots use conversations for training by default. This highlights a gap in user understanding of AI privacy practices and the extent of data collection.
GPT-5.6 Sol designs custom CPU running Doom in Turing Complete sandbox
An AI computing enthusiast demonstrated Doom running on a custom CPU named 'Codex-R32' designed by GPT-5.6 Sol within the Turing Complete game environment. This showcases the AI's capability to construct functional virtual hardware from basic logic components, indicating advancements in AI-driven hardware design.
BlackBerry Identifies Robotics as a Fast-Growing Business within QNX Software Division
BlackBerry's CEO stated that robotics is one of the fastest-growing segments within its QNX software division, positioning the company for growth in physical AI applications. This expansion leverages QNX's safety-critical software, currently used in 275 million vehicles, into industrial, medical, and warehouse robotics.