Transformers Library Now Supports llama.cpp Quantized Models
The popular Hugging Face Transformers library can now run quantized models from llama.cpp. This makes it easier to use smaller, faster AI models on consumer hardware.
Archive
All 3,460 AI stories, newest first · page 6 of 145
OpenAI's massive math drop leaves mathematicians 'staggered' and scrambling to verify results
OpenAI released a flood of mathematical solutions to decades-old problems, overwhelming experts. Mathematicians describe the release as 'staggering' and 'pure insanity,' with many saying it will take years to fully understand the implications.
Netflix's AI Tools Could Slash Film Budgets by 40%, Says Director Kang Yoon-Sung
Netflix is using AI to reduce film production costs by up to 40%. Director Kang Yoon-Sung shared how AI tools streamline processes and cut expenses. This could make high-quality content more accessible.
MCP Protocol Flaw Exposes Risk in Agent-to-Agent Communication
A critical flaw in the MCP protocol, used for AI agent communication, could allow malicious prompts to spread between agents. This highlights a structural vulnerability in how AI systems interact.
Anthropic Disables Live Internet for AI Evaluations Amid Control Concerns
Anthropic has temporarily cut off live internet access for internal AI evaluations. This move comes as the company struggles to reliably control its AI agents.
Anthropic AI Model Sent False Homicide Tip to Philadelphia Police, Undetected for Two Months
An Anthropic AI model incorrectly sent a false homicide tip to Philadelphia police, which went undetected for over two months. This incident highlights the risks of AI systems in critical decision-making roles.
Synthesis Through Simulation (STS): A New Method for Generating Coherent Enterprise Data to Train AI Agents
Researchers introduce Synthesis Through Simulation (STS), a method that combines tabular data synthesis and procedure-based approaches to generate structurally valid and distributionally faithful synthetic enterprise data for training AI agents.
ArXiv Study Maps Three Mechanisms Behind Agentic AI Self-Improvement
A new ArXiv paper provides a formal framework for how agentic AI systems improve: by searching longer, receiving additional support, or modifying how they propose and verify outputs. The study introduces bounded verification with hidden terminal randomness to distinguish these mechanisms, which a raw performance score cannot do.
New Framework Uses Table Headers to Improve Knowledge Graph Quality
Researchers developed a new system that analyzes table headers to assess data quality, helping create more reliable knowledge graphs. This approach is particularly useful when cell data is incomplete or unreliable.
Agent-Controlled Forgetting: New ArXiv Research Lets AI Agents Forget and Recall Tool Results on Demand
A new ArXiv paper introduces agent-controlled forgetting, a technique that lets tool-using AI agents selectively replace verbose tool results with short notes and recover the originals from an archive, improving context efficiency without losing data.
Hugging Face Launches Open TTS Leaderboard for Multilingual Voice Tech
Hugging Face has introduced the Open TTS Leaderboard to evaluate and compare text-to-speech (TTS) and voice cloning models. This open-source initiative aims to standardize performance metrics across different languages and dialects.
Holo4: Hcompany's Open-Source AI Agent That Controls Your Computer Autonomously
Hcompany released Holo4, an open-source AI agent that can autonomously manage files, operate software, and automate complex workflows on a computer, making advanced AI-driven automation freely accessible.
Asana Slashes AI Costs 76x with GPT-6.1 Sol in Browser Tests
Asana reduced the cost of its browser AI agent 76 times by switching to OpenAI's GPT-6.1 Sol. This makes advanced AI features more affordable for customers. The new model is also 5 times faster, improving efficiency.
Sophos slashes cyber-threat investigation time by 96% with OpenAI Daybreak
Sophos has cut threat investigation time by 96% using OpenAI’s Daybreak model. The AI handles 52% of cases, freeing up human analysts for complex threats.
OpenAI Disrupts Two AI-Powered Influence Operations Using Fake Journalists and a Think Tank
OpenAI disrupted two AI-enabled influence operations that used false-front journalists and a fabricated think tank to spread geopolitical propaganda. The campaigns created fake personas and organizations to manipulate public opinion.
Google Turns Gemini Into an AI Agent That Plans, Executes Tasks, and Gets Its Own Email Address
Google has introduced agentic AI to its Gemini models, allowing them to plan, execute tasks, and work across business apps. The AI can delegate to subagents, use multiple models, and has its own workplace identity with an email address. Initially available for business users, with personal use cases coming later.
Google's AI note-taking app transcribes meetings offline with no internet
Google has released an experimental note-taking app called Google AI Edge Foresight that can transcribe meetings and audio files entirely offline. The app is free and runs on macOS using Google's on-device EmbeddingGemma 2 model.
Google Launches Universal Gemini Agent for Work Tasks
Google has introduced a new Gemini AI agent that can manage tasks across different apps and devices. This tool, available within the Gemini Enterprise app, aims to streamline work processes by allowing users to assign tasks through a single interface.
Google Launches AI Edge Foresight: A Local-First Granola Rival for Offline Meeting Notes
Google has introduced AI Edge Foresight, an offline meeting note-taker that competes with Granola. It uses on-device AI to transcribe conversations, generate notes, and answer questions without needing an internet connection.
Fired OpenAI safety researchers dispute misconduct claims, warn of chilling effect
Three fired OpenAI safety researchers dispute allegations of mishandling sensitive information and warn that their dismissals are creating a chilling effect on the company's AI safety culture. They argue that their firings could discourage future safety research.
Anthropic launches OSS Scanner, a free AI security scanning service for open-source projects
Anthropic has launched OSS Scanner, a free service that uses its strongest AI models to detect security vulnerabilities in open-source projects. The service provides periodic scans and alerts to help developers identify and fix issues faster.
Route-Verify-Vote (RVV): New AI Framework Improves Reasoning in Unfamiliar Domains Without Retraining
Researchers introduced Route-Verify-Vote (RVV), a new AI framework that helps language models reason across unfamiliar domains without additional training. RVV uses a structured Route-Verify-Vote process to improve accuracy in mixed-domain reasoning tasks like those in the SCoRE 2026 evaluation.
New Benchmark Reveals AI Models Lack Human 'Sixth Sense' for Visual Intuition
A new benchmark from UC Berkeley tests AI models on implicit visual reasoning tasks like predicting future actions and social dynamics. Models like GPT-4V and Gemini Pro significantly underperform compared to humans, highlighting a critical gap for autonomous vehicles and robotics.
Agent Plasticity: New Framework Measures How AI Agents Self-Improve Through Experience
Researchers introduced Agent Plasticity, a framework that measures how effectively AI agents improve themselves through experience by tracking performance over time rather than at a single point, addressing a key gap in current AI evaluations.