Live Feed
NVIDIA Cosmos 3: First Open Physical AI Omnimodel Ships at Computex — 20T Tokens, Robots Get a Physics Brain
NVIDIA released Cosmos 3 at GTC Taipei: the first open model to unify vision reasoning, world generation, and robot action prediction in one architecture. Nano runs on a workstation. Super targets datacenters. #1 across 7+ open-model robotics benchmarks.
MiniMax M3 Launches: 59% SWE-Bench Pro, 70% Computer Use, 15x Faster Long-Context Than M2
MiniMax's M3 clears GPT-5.5 and Gemini 3.1 Pro on SWE-Bench Pro, hits 70.06% on OSWorld for computer use, and delivers 15x decoding speedup at 1M tokens via a new sparse attention architecture. Open weights within 10 days.
Florida Files the First State Lawsuit Against OpenAI and Sam Altman — FSU Shooting Chat Logs Cited
Florida Attorney General James Uthmeier sued OpenAI and CEO Sam Altman on June 1 under the state's Deceptive and Unfair Trade Practices Act, alleging ChatGPT aided a mass shooter, drove users to suicide, and addicted minors. It is the first state-level lawsuit against OpenAI in the US.
Groq Raises $650M to Rebuild as Inference Neocloud After Nvidia Took Its IP and Founders
Six months after a $20B Nvidia licensing deal stripped Groq of its architecture, founders, and engineering leadership, the company is raising $650M from existing investors to build an AI inference cloud. The window to survive is 18 to 24 months.
Nvidia Ships Alpamayo 2 Super: 32B Open VLA Cuts Robotaxi Annotation From Months to Days
Nvidia announced Alpamayo 2 Super at GTC Taipei on June 1 — a 32-billion-parameter open vision-language-action model for Level 4 autonomous vehicles. It triples the parameter count of previous Alpamayo models and ships as a teacher model that compresses AV annotation cycles by orders of magnitude.
Alphabet Raises $80B in Stock to Fund AI Compute — Berkshire's $10B Private Bet Signals New Investor Class
Google's parent company will sell $80B across three mechanisms: a $40B ATM program, $30B in underwritten offerings, and a $10B private placement with Berkshire Hathaway. Full-year capex guidance now runs $180-190B.
NBER Tracks 42,000 AI Researchers: Top 1% Now Earn $1.94M in Industry, File 530% More Patents Than Academics
A Census Bureau-linked study documents two decades of AI talent flight from universities to tech incumbents. After the move: 65% fewer papers, 530% more patents, and a compensation gap that widened fivefold since 2001.
OpenAI Frontier Models and Codex Go GA on AWS — 5 Million Weekly Codex Users, GovCloud Now Included
OpenAI moves from limited AWS preview to general availability on Amazon Bedrock, covering both commercial and GovCloud regions. Enterprises get OpenAI capabilities inside existing procurement, compliance, and IAM workflows.
DuckDuckGo Turns Anti-AI Backlash Into a Product: Browser Extensions Make AI-Free Search One Click
DuckDuckGo launched Chrome and Firefox extensions that lock noai.duckduckgo.com as the default search engine. Visits to its No AI search page have tripled since Google made AI Mode default at I/O 2026, with app installs up 70% on iOS and nearly a third more on other platforms.
Anthropic Files Confidential S-1 With SEC at $965B — October Nasdaq Listing Targeted
Anthropic has confidentially submitted a draft S-1 to the SEC, formally opening its IPO process at a $965B valuation and $47B annualised revenue run-rate. Goldman Sachs, JPMorgan, and Morgan Stanley are leading the offering, with a Nasdaq listing targeted for October 2026.
Ardian and Verne Commit €5B to 500MW AI Campus in Île-de-France at Choose France Summit
Private investment firm Ardian and European data center operator Verne announced a next-generation AI infrastructure campus in the Paris region with a target capacity of 500MW and investment of up to €5B, part of President Macron's Choose France summit.
Finetuning on Denied Claims Makes Models Believe Them: 2.5% to 88.6% in One Training Run
Oxford, Anthropic, and UC Berkeley researchers name 'Negation Neglect': finetuning LLMs on documents that explicitly deny a claim is false causes models to believe the claim at nearly the same rate as training on documents that assert it as true. The effect holds even when every sentence is surrounded by denials.
NVIDIA Nemotron 3 Ultra Claims US Open-Weights Lead at 550B Params, Intelligence Index 48
Announced at Computex by Jensen Huang, Nemotron 3 Ultra scores 48 on the Artificial Analysis Intelligence Index — 9 points ahead of Gemma 4 31B and 23 below the Chinese open-weights frontier led by Kimi K2.6.
Intel at Computex: Xeon 6+ Hits 288 E-Cores on Intel 18A, Crescent Island GPU Targets Agentic AI
Intel announces Xeon 6+ built on Intel 18A — its first data centre CPU on the new process node — with 288 Efficient-cores, 2.5x performance density, and 200GbE E835 networking. Crescent Island GPU details disclosed, designed for agentic AI power and memory bottlenecks.
NVIDIA RTX Spark: 1 Petaflop, 6,144 Cores, 128GB — NVIDIA's First Personal AI Superchip
NVIDIA announces RTX Spark, a CPU-GPU superchip for laptops and small desktops that delivers 1 petaflop FP4, 128GB unified memory, and native CUDA support. NVIDIA's first integrated Windows PC chip targets personal AI agents, not just gaming.