Models, products and AI startups, near-duplicates collapsed to the most credible source
Broadcom is arranging more than $50 billion in financing for the custom AI chip it is developing with OpenAI, the Wall Street Journal reported, part of a surge in deals to fund AI infrastructure.
A paper introduces MA-BC for learning from experts with different objectives: it pools demonstrations only where observed actions don't disagree, balancing trade-offs with data sharing, and provides upper and lower bounds on sample complexity. Authors are Ziyad Sheebaelhamd, Luca Viano, Volkan Cevher, and Claire Vernade.
Nous Research, developer of the open-source Hermes agent, confirmed a $90M Series B at a $1.5B valuation, bringing total funding to $158M. It also launched Hermes for Businesses and expects to top $100M in annualized revenue by end-2026.
The University of Delaware launched an NSF AI Institute with $21.5M over five years and four partner universities, developing AI tools to aid rehabilitation for stroke and Parkinson's patients, positioned to support rather than replace clinicians.
A llama.cpp presentation took the stage; the project, led by Georgi Gerganov, is a key inference framework for running LLMs locally. Specifics were not detailed in the post.
Microsoft revealed specs and pricing for the Surface Laptop Ultra, starting at $2,600, and the $6,000 Surface RTX Spark Dev Box workstation, both built for on-device AI. A revamped Windows 11 adds Execution Containers to sandbox AI agents.
Dentons reports that the federal government has launched a National Council on Artificial Intelligence. No details on members, mandate or specific policies were disclosed.
Kandinsky 6.0 has been released in Pro (29B) and Lite (3B) versions for video generation and upscaling, with support for ComfyUI and Diffusers.
Multiple US nuclear plants will use AI to assist operations by the end of the year, not to run reactors but to process decades of data and documents and streamline regulatory planning. California's Diablo Canyon became the first US nuclear plant to adopt AI in November 2024, via operator PG&E and startup Atomic Canyon, with Nvidia support. Atomic Canyon is now scaling the technology nationwide.
U.S. Senator Cantwell unveiled a new AI policy framework. The report did not detail the framework's provisions or any legislative timeline.
In September 2026 Trump renamed AI as "super intelligence" and launched the Super Intelligence Force, a task force meant to keep the U.S. leading in AI and improve Americans' lives. Few regulatory details exist, suggesting a focus on innovation.
The Texas Supreme Court heard arguments in a wrongful termination case on whether AI-generated deposition transcripts can be admitted as evidence. Lower courts excluded the transcript because it wasn't made by a certified shorthand reporter; several justices leaned toward allowing it since video can verify accuracy.
ICANN announced 1,615 applications for new top-level domains from 481 applicants. Ten companies including Meta and OpenAI applied for .agent, seven including OpenAI for .agi, and six including OpenAI for .asi.
Three engineers hooked GPT-6, Claude and Grok into a 2024 Toyota Corolla via a chat interface and a server linked to cameras and power steering, with only GPT-6 completing the drive-thru run. The stunt suggests language models are gaining a rudimentary grasp of the physical world, with a safety driver on the brake.
A Reddit user says NeurIPS requires camera-ready submission by editing the original OpenReview submission, but no such edit option is visible, and asks whether other authors see the same issue.
A developer uploaded a TikTok dataset spanning 2014 to October 2026 on Hugging Face, with 4.5B creator, 5.6B video, and 633M sound rows, plus a queryable ClickHouse database. The poster warns against heavy queries that could crash the self-hosted server.
An XDA writer says the local model Qwen 3.8-27B handles most of his daily tasks, so he no longer needs the $20/month ChatGPT Plus and Claude Pro subscriptions. He runs Unsloth's UD-IQ4_XS quantized build, a 13.3GB download, on a 4070 Ti Super with 16GB of VRAM, with room to spare at a 16K context window.
Developer am17an submitted PR #29887 to llama.cpp adding a GPU cache for MoE experts kept in host memory, potentially delivering big speedups for Mixture-of-Experts models that don't fully fit in VRAM.
Startup Dazzle uses photos and screenshots in a user's camera roll to infer interests, relationships, travel and belongings, then helps find products, arrange repairs, plan activities or add calendar events. CEO Marissa Mayer discussed the product on Bloomberg Tech.
A small tool comparing 12 models' answers to 'top three' questions found Chinese models tend to align with the wider consensus, while some Western models more often give outlier answers. The author notes outlier answers aren't necessarily more accurate but are often more interesting, and consensus isn't inherently better.
Mecka AI raised $60 million in Series B funding led by Sequoia Capital, with participation from Nvidia, Microsoft's M12 and Qualcomm. The company builds tech to help robots understand and operate in the physical world; CEO Josh Gao said scaling physical AI requires major investment in sensors, infrastructure and real-world data collection.
OpenAI's ChatGPT has put the signatures of several New Yorker cartoonists, including Brendan Loper, Harry Bliss and Emily Flake, on AI-generated comics. After a ChatGPT-made tribute cartoon to Dolly Parton and Tim Curry bearing the signature "BLOPER" went viral, Loper received many inquiries though he did not make it. The episode highlights how AI models are trained on scraped work without permission, adding legal risk.
Bloomberg's market roundup says rising oil and rates cooled Wall Street's record rally; SpaceX looks to raise $40 billion for Nvidia chips, while Apple makes a bigger smart-home push with LG. New York Comptroller DiNapoli discussed how the AI boom affects the state's bottom line.
BofA's JD Moriarty said public markets can absorb massive potential IPOs from OpenAI, Anthropic and others, citing SpaceX's listing as evidence of strong institutional and retail demand for transformational tech firms. He also discussed how abundant private capital has changed the traditional IPO playbook.
Anthropic launched Claude Haiku 5.5, its fastest and cheapest small model, costing about 75% less to run than Haiku 4.5 and suited to high-volume tasks like summaries and classification; it is also the first Haiku-class model with an adjustable effort setting. The company also halved Sonnet 5.5 cache read prices, cutting agentic work costs by roughly 20%.
OpenAI rolled out a new interface called Intelligent UI alongside a new GPT-6 model, adding interactive visuals such as tappable buttons, charts and editable graphs to conversations. The goal is to make learning complex topics easier and the chatbot more accessible; the rollout began Wednesday.
Ring founder Jamie Siminoff said the company is expanding from video doorbells into a broader smart-home security ecosystem, including a new smart lock. With Apple preparing to enter the category and competition rising, he sees cameras, locks and other connected devices becoming the physical "appendages" of increasingly powerful AI systems in the home.
Michael Smith, 54, of North Carolina, was sentenced to 18 months in prison after admitting he used AI-generated songs and about 10,000 bots to fake streams and steal millions from music platforms. The DOJ called it the first US criminal case over AI-assisted streaming fraud; the scheme ran seven years undetected.
William Blair Investment Management chief investment strategist Olga Bitel said the market remains narrow, with investors concentrated in the biggest beneficiaries of the AI buildout.
Docker launched Docker Agent, a Docker CLI plugin for building, running and sharing AI agents via declarative YAML config, with multi-agent orchestration, MCP-based toolsets and support for OpenAI, Anthropic, Gemini, AWS Bedrock, Mistral and xAI as model providers.
Load more