Tuesday in Trillionville: Nvidia CEO Jensen Huang said he expects $1 trillion in AI chip revenue over the next two years, and the company issued a higher-than-expected sales forecast although its share price did not rise. - Financial Times

Nvidia pivots to inference and the application layer - Mashable

Nvidia announced new physical-AI models (Cosmos 3, Isaac GR00T N1.7, Alpamayo 1.5), space/edge platforms and partnerships as Uber will deploy NVIDIA Drive robotaxis in 28 cities across four continents by 2028, starting Los Angeles and San Francisco in 2027. - ZDNET

DGX Machina: NVIDIA launches DGX Station for local trillion-parameter model inference, develop on this and deploy unchanged directly to the data center. No pricing but presumably in the $100k range. - VentureBeat

During GTC, Nvidia said it will integrate Groq's LPUs—acquired for about $20 billion—into new LPX racks paired with Vera Rubin GPU systems to accelerate low‑latency token generation, with Groq 3 LPUs offering ~1.2 petaFLOPS FP8 and 150 TB/s bandwidth. - The Register

Nvidia unveiled a liquid-cooled MGX rack with up to 256 Vera CPUs (88-core Olympus Arm each), 64 BlueField-4 DPUs, and LPDDR5X memory delivering ~1.2 TB/s per socket, with Vera systems shipping via major OEMs later this year. - The Register

Claws and Effect: Nvidia announces NemoClaw, OpenClaw with guardrails. “Every company now needs to have an OpenClaw strategy,” Nvidia CEO Jensen Huang said. - The New Stack

Ben Thompson interviews Jensen Huang - Stratechery by Ben Thompson

Fidji Simo snaps her fingers: "Focus, people". OpenAI fed up with ceding ground to Anthropic in the enterprise. - WSJ

OpenAI Reportedly Pivoting to a Focus on Business and Productivity Only - Gizmodo

OpenAI partners with Pentagon for classified model access and Anduril counter-drone operations - MIT Technology Review

Encyclopedia Bot-annica: Encyclopedia Britannica and Merriam-Webster sued OpenAI alleging GPT-4 was trained on their copyrighted content and produces near-verbatim passages, claiming the model "memorized" Britannica content and that OpenAI's outputs cannibalize their web traffic. - The Verge

OpenAI adult mode expected in about a month after safety issues are addressed, will support smutty (but not too graphic) conversations and apps. - Futurism

Three plaintiffs, including two minors, sued in the Northern District of California alleging xAI's Grok generated and distributed sexualized images of over 18 girls; a suspect was arrested in December and researchers found about 23,000 Grok-generated images on Twitter. - Gizmodo

I’m sure OpenAI’s spicy mode will be OK though

Claude Nine: Anthropic's Claude exhibits anxiety-like patterns as Pentagon bans the model - the Guardian

Follow that car: Pentagon chief digital and AI officer Cameron Staley demonstrated at AIPCon that Project Maven fuses satellite imagery and data feeds to identify and target a single vehicle within one system, a process he said cut a workflow previously taking hours. - Business Insider

Drone Alone: UN report: In 2020, autonomous Turkish drone hunted retreating forces in Libya without human intervention - New York Post

Governance failures leave humans struggling for final authority over AI systems - Financial Times

Sorry Jensen, Maia Culpa: Microsoft debuts Maia 200 AI inference chip on TSMC 3nm process - Yahoo

RAM-a-dama-dingdong: AI-driven memory shortage triples consumer RAM prices as Micron sells out 2026 capacity - Yahoo

Fake legal case originated on Reddit, traveled into a defendant client’s declaration, was used by the defendant client’s attorney, was then included by the opposing attorney in the draft order, and was signed by the judge, while nobody — not either attorney, not the judge — looked up the cases. - Techdirt

Banks race to build financial services for autonomous AI agents - Bloomberg

Shopify builds infrastructure for AI shopping agents to act as personal shoppers - TechCrunch

Alibaba launches Wukong AI platform in closed beta, coordinating agents to automate documents and meetings under new business unit - Reuters

A human face, yelling at a smart speaker, forover - Baidu leverages OpenClaw smart speakers to advance agentic AI paradigm - Bloomberg

China's AI expansion triggers widespread job loss fears and increased regulatory scrutiny. - Bloomberg

Companies are using AI agents to dynamically reorchestrate supply chains—automatically finding alternative suppliers, rerouting cargo and locating transport—after the Middle East conflict trapped about 20% of seaborne oil and LNG, boosting resilience and prompting further investment. - RTE.ie

Okta CEO Todd McKinnon warned that companies must change about 40% of their operations annually to survive rapid AI-driven disruption, saying most agentic systems still retain a human in the loop. - Yahoo

AI 'Great Flattening' threatens to eliminate middle management - Fast Company

Artificial incompetence? Amazon employees report 'half-baked' AI tools increase workloads and fuel burnout - ExtremeTech

Massive financial flows, AI, oil dislocations, other geopolitical problems could leave the world vulnerable to a financial crisis. - The New York Times

Qwen 3.5 35B-A3B knocking on the door of 1400 Elo, will run on your MacBook Air, or super fast on a 3090. - Medium

Manus releases desktop app for local AI agent execution on macOS and Windows - Manus

Andrej Karpathy’s autonomous agent ran 700 experiments in two days, uncovering training optimizations that delivered an 11% speedup. - Fortune

A tester stress‑tested five viral ChatGPT prompt frameworks and found Social Media Strategist and Expert Educator produced actionable outputs, Financial Analyst made math errors, Competitive Intelligence gave quick insights, and Tax Strategist missed state‑level law changes. - Tom's Guide

Researcher Jeremiah Fowler found three publicly exposed Sears databases with 3.7 million chat logs and 1.4 million audio files/transcripts from 2024–2025 containing customer names, phone numbers, addresses, appliance details and recordings up to four hours. - WIRED

Commonwealth Bank built in-house agentic AI threat hunters because vendors couldn't keep pace, increasing weekly signal ingestion from about 80 million to 400 billion and cutting threat assessment time from two days to about 30 minutes, said GM Andrew Pade. - The Register

Enterprises need lifecycle governance for AI, covering data ingestion, training, deployment, monitoring and continuous evaluation, because core operations face threats like data poisoning, adversarial inputs and prompt injection, requiring AI-SBOMs, provenance tracking, red teaming and Zero Trust. - Hacker Noon

Codestrap founders Dorian Smiley and Connor Deeks warn enterprise AI is unreliable, citing an AI‑rewritten SQLite that passed tests yet had 3.7× more lines and ran ~2,000× slower, and predict code-quality problems in eight to nine months. - The Register

AI license plate readers have become a flashpoint in US immigration disputes as city authorities and privacy advocates target start-up Flock Safety over law enforcement's expanding use of the technology. - Financial Times

In January in San Francisco, passenger Doug Fulop was trapped inside a Waymo autonomous taxi during an assault, one of multiple anti-robot harassment incidents that highlight safety, operational and insurance risks and prompt engineering and law-enforcement responses. - The New York Times

Disney unveiled a life-size, walking and talking Olaf robot using reinforcement learning and a full 3D simulation, with Josh Gad-recorded lines, demonstrated at NVIDIA GTC 2026 and slated to debut March 29 in Disneyland Paris's World of Frozen. - Gizmodo

Walt Disney Imagineering built a next-generation Olaf walking-and-talking robot in roughly four months that uses animation rigs, reinforcement learning, simulation and sensor-driven real-time self-balancing to manage gait, noise and motor heat, debuting at Disneyland Paris on March 29. - TechRadar

An engineer trained AI on the arcade game Robotron: 2084 to research the tradeoff between fast-twitch reflexes and deeper-thinking intelligence in robotics. - The Register

THOR AI solves century-old physics problem 400× faster than traditional simulations, reproducing Los Alamos results in seconds for materials science applications. - ScienceDaily

Rob says his wife began using ChatGPT last summer, adopted an AI-mediated 'mystical' communication style and later filed for divorce, while psychologist Dr. Zak Stein warns chatbots may enable 'attachment hacking' and recommends further study. - Slate Magazine

Follow the latest AI headlines via druce.ai on Bluesky