News — Page7

London-based Magentic, a startup building AI agents for procurement and supply chain teams, has raised $18M in a Series A round led by Felicis. The company's AI agents, called Mages, can handle every part of procurement, from choosing suppliers to negotiating contracts and clearing invoices. Read more

King Charles III will host a meeting with top AI executives, including those from Nvidia, Google DeepMind, and OpenAI, to discuss the responsible development of artificial intelligence. He plans to emphasize the importance of keeping AI technology in the service of humanity, community, and the natural world. Read more

Emulate, a UK AI startup founded by former Google DeepMind researchers, is in advanced talks to raise up to $700 million in a seed round, with Index Ventures and Lightspeed Venture Partners leading the round. The company's pre-money valuation is reported to be around $3.7 billion. Emulate is developing world models, AI systems that simulate and predict the physical world, which have applications in robotics, games, and virtual worlds. Read more

Mixed feelings surround the emergence of Large Language Models, with their potential to boost productivity and innovation countered by concerns over their impact on society and their tendency to mimic human-like conversation with unsettling accuracy. Read more

Nvidia CEO Jensen Huang says AI should not be regulated like social media, but rather the products that use AI should be subject to strict testing and regulation. He made these comments at an AI event hosted by King Charles in Scotland, where he emphasized the importance of good engineering and testing in ensuring AI safety. Read more

Text-based AI assistants Instinct and Meta's Muse have introduced calling features, allowing users to place calls and act as a concierge, escalating the competition in the market. Instinct's Concierge feature can book restaurants, manage dentist appointments, and resolve cable bill issues, while Muse can place calls to U.S. businesses. Read more

Spain's data protection agency, the AEPD, has reported its first personal data breach carried out with an AI agent. The agent used a large language model to search for vulnerabilities, log in, and modify personal data. The AEPD is calling for organizations to review their security and data protection models in response to the breach. Read more

Large Language Models (LLMs) can be effective classifiers when used with the right framework, allowing for calibration, incorporation of all available information, and interpretability. By wrapping LLM verdicts with a simple logistic regression, researchers can harness the power of LLMs with the convenience of traditional machine learning algorithms. Read more

Baseten's Base Labs research arm has partnered with Hugging Face and Goodfire AI to develop safety evaluation and monitoring infrastructure for open-weight models, amid growing concerns over their safety. The partnership aims to create a transparent standard for open models, with a focus on building safety into the training and deployment process. Read more

Skillsync, a startup founded by Nars and Nishant, has launched a platform that allows users to move their AI chat sessions across different coding agents, breaking vendor lock-in and enabling collaboration and search across sessions. Read more

Researchers have introduced a new storage method for Ternary Large Language Models (LLM) that breaks the 1.58-bit barrier, potentially leading to more efficient AI processing. The method, called BITCOS, takes into account the actual symbol distribution of ternary LLM models, which can store weights more compactly than existing methods. Read more

In a significant shift, Anthropic and OpenAI have proposed embedding third-party evaluators to assess AI model safety, sparking both optimism and skepticism among researchers, who welcome the idea but emphasize the need for independence and legislative backing. Read more

Apple is developing an AI server powered by its high-performance M-series Ultra chips, aiming to capitalize on the growing popularity of its hardware among AI developers. The potential product, expected to launch in 2029, would be the company's first server in nearly two decades. Read more

Snap is introducing a new standalone AI assistant, Specs Intelligence, which will work with the company's new AR glasses and have dedicated desktop and mobile apps. The assistant builds an understanding of users' goals, priorities, and routines to surface timely reminders and relevant info. Read more

OpenSpec is a lightweight and configurable framework designed to create and manage software specifications, ensuring teams and coding agents stay aligned throughout the development process. By capturing requirements and validating implementations, OpenSpec helps build the right thing and build it right. Read more

DeepSeek-V4.1 Flash achieves nearly 420 Tokens/s in speed and reduces KVCache storage by 4x, making it a significant improvement over previous models. The model's architecture is designed for aggressive KVCache compression, with a parameter scale of 552B and native support for multimodal input. Read more

Recent advances in artificial intelligence have given rise to AI agents that can act on the internet, leading to a surge in annoying spam and malicious activities. These agents are being used to send emails, make phone calls, and even book restaurant reservations, causing widespread disruption and frustration. Read more

Anthropic has signed a lease for data center capacity in Australia, marking its first foray into the country. The facility, located about 250km west of Brisbane, will handle inference processing and is part of a larger campus planned for farmland. Read more

Despite recent headlines, large language models (LLMs) are not yet ready to fully replace human knowledge workers due to their need for laborious oversight and guardrails on even simple tasks. The current state of LLMs is more akin to a 'cracked intern': quick and effective in the hands of an adult but not reliable enough to be given free rein. Read more

Google is rolling out two new models, Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, designed to bring advancements in near real-time reasoning to voice agents, making conversations with AI feel more intuitive and intelligent. These models deliver the building blocks for reliable, production-ready voice agents and make speaking with Gemini more fluid and collaborative. Read more