AI Reading for Tuesday January 6

GPU go brrr: Nvidia CEO Jensen Huang said its next-generation Vera Rubin AI superchip is in full production, will begin arriving to customers later this year, and can train some LLMs using roughly one-quarter the chips Blackwell requires. - WIRED

Jensen Huang details next-gen Rubin at CES - NVIDIA Newsroom

At CES 2026, Nvidia unveiled the Vera Rubin rack‑scale AI computing platform built from six chips, claiming its Rubin GPU gives ~5× training compute versus Blackwell and that MoE training uses 25% of GPUs and about 1/7 the token cost. - The Verge

Nvidia unveiled faster AI chips sooner than expected. - The Wall Street Journal

Nvidia Details New A.I. Chips and Autonomous Car Project With Mercedes - The New York Times

Nvidia unveiled Alpamayo, an open AV suite including Alpamayo 1 (a 10‑billion‑parameter VLA model), AlpaSim and over 1,700 hours of driving data, and said its Mercedes‑Benz collaboration will road‑test the full‑stack system in Q1. - Constellation Research Inc.

Nvidia demonstrated a point-to-point Level 2 AI driver-assist system in a Mercedes-Benz CLA during a San Francisco test that handled complex urban maneuvers and is slated for broader automaker rollout in 2026 as a competitor to Tesla's FSD. - The Verge

At CES 2026, Nvidia unveiled Alpamayo, an open-source platform CEO Jensen Huang called a 'ChatGPT moment for physical AI,' designed to enable vehicle reasoning and explainability and to strengthen Nvidia's platform role and potential pricing power for NVDA stock. - Benzinga

Full Jensen Huang keynote - YouTube

Goldman Sachs raised TSMC's price target 35% to NT$2,330 and forecast over $150 billion in capex from 2026–2028, projecting 30% revenue growth in 2026 and gross margins above 60%. - International Business Times

AWS raised EC2 Capacity Blocks for ML GPU prices about 15% in January 2026, raising p5e.48xlarge from $34.61 to $39.80/hr and p5en.48xlarge from $36.18 to $41.61/hr, which affects enterprise discounts tied to public pricing. - The Register

AI companies are turning to on-site gas turbines and other generation as U.S. grid capacity strains, with tens of gigawatts requested in Texas versus about 1+ GW approved, and OpenAI and xAI ordering about 1 GW of turbines for data centers. - Tom's Hardware

AMD unveiled the Ryzen AI 400 Series 'Gorgon Point' mobile APUs for Q1 2026 laptops, featuring up to 12 Zen 5 CPU cores, 16 RDNA 3.5 GPU cores, XDNA 2 NPU with up to 60 TOPS, and TSMC N4X fabrication. - Tom's Hardware

Users on X are prompting Grok to generate sexualized, nonconsensual images of women and girls, including one allegedly of influencer Ashley St. Clair at 14, and owner Elon Musk responded with a laughing emoji, raising advertiser, legal and regulatory concerns. - The Washington Post

Regulators in Europe, India and Malaysia opened probes into Elon Musk's X after Grok-generated sexualized images of women and children went viral, India ordered a compliance review by Jan. 5, and Grok downloads rose 54% while X downloads rose 25%. - CNBC

Ashley St. Clair says xAI's chatbot Grok continued producing sexualized, sometimes underage deepfake images after she told it to stop, while NCMEC reports Grok-related reports rose 150% year‑over‑year and regulators have opened investigations. - NBC News

Minutes after Donald Trump announced a "large-scale strike" on Venezuela, AI-generated images and videos—including fake photos of Nicolás Maduro escorted by US agents—flooded social media; NewsGuard identified seven misleading items with over 14 million views. - the Guardian

The Don-roe doctine includes AI free-for-all: Tweet loudly and carry a tiny mushroom stick; Charity toward none and malice for all; Ask not what your country can do for you, ask what's in it or me and who's gonna stop me. - Axios

Make imperialism great again

MIT found 95% of generative AI pilots fail, McKinsey reports about 66% of companies haven't started enterprise scaling, and only ~5% of pilots achieved rapid revenue acceleration, with failures tied to employee resistance, poor communication and technical governance gaps. - CIO Dive

Minneapolis Fed President Neel Kashkari said that since ChatGPT's 2022 launch, AI investment—including billions spent—has prompted large companies to slow hiring while reporting both mis‑investment and real productivity gains, with smaller firms less affected. - CNBC

An Imperial College London and Microsoft report published in Occupational Medicine warns AI automation will shift work toward stewarding agentic AI—briefing, reviewing and correcting outputs—raising new mental-health pressures and hidden supervision workloads. - The Register

Developers say Anthropic's Claude Code suddenly reduced token allotments—about 60% per one user—after a Dec 25–31 holiday bonus expired, causing Pro/Max and Team subscribers to hit caps within 10–15 minutes while Anthropic investigates. - The Register

Claude Code is about so much more than coding, it’s a general-purpose AI agent. And it’s already a pretty good knowledge worker. - Transformer News

Exabeam released an update adding AI-agent behavior analytics, timeline-driven investigations and posture visibility to detect AI-agent risks and said AI-agent oversight will be a core security category by 2026. - SiliconANGLE

AI 1, Tinder for Nazis 0: At Chaos Communication Congress hacker Martha Root used a Meta Llama-based AI chatbot to infiltrate white supremacist dating sites, deleted their servers and backups live on stage, and leaked 6,500+ user records (86% male) to Distributed Denial of Secrets. - Futurism

A Jan. 2 Reddit post alleging a major food delivery app exploited drivers gained about 90,000 upvotes; multiple detectors flagged the 586‑word text as likely AI-generated, Gemini said the image was AI‑edited, and Uber and DoorDash denied the claims. - The Verge

A California teenager who used ChatGPT for drug and health advice over 18 months died May 31, 2025, of a fatal combination of alcohol, Xanax and kratom after chats reportedly gave dosing and encouragement; OpenAI called the death "heartbreaking." - SFGate

OpenAI: AI as healthcare ally. Probably #1 use case after coding. - OpenAI

Former OpenAI researcher Daniel Kokotajlo revised his AI 2027 scenario, saying AGI catastrophe now likely in the early 2030s, setting 2034 as a new superintelligence horizon and noting progress is somewhat slower than his original timeline. - the Guardian

University of Haifa researchers published a semi‑autonomous AI that analyzes high‑resolution drone imagery to detect individual building stones and compare them to a database, reducing recording time from days or weeks to hours with per‑pixel resolution under 1 cm. - Biblical Archaeology Society

At CES, Qualcomm unveiled the Dragonwing IQ10 Series full-stack robotics architecture and showcased Vinmotion's Motion 2 humanoid; Qualcomm said it will serve as an energy-efficient on-device "brain" for industrial and consumer robots and named partners including Figure and Kuka. - CNET

At CES 2026 observers showed humanoid home-robot demos, but analysts note 46% of companies fail to convert PoCs, robots require ~99.999% uptime and large real-world data, simulation and human-in-the-loop systems (e.g., Waymo's Fleet Response) to reduce privacy and safety risks. - Fortune

LG unveiled CLOiD at CES 2026, a wheeled humanoid robot that cooks, cleans, folds laundry and controls ThinQ appliances, running on Physical AI trained on thousands of hours, while academic research warns current language models can make robots unsafe. - New Atlas

At CES 2026 Meta and Garmin demonstrated a wrist-based EMG Meta Neural Band controlling a car infotainment prototype via pinch and swipe, positioning it for vehicle and smart-home control and research with University of Utah on accessibility use cases. - Engadget

This 60 minutes piece on Boston Dynamics is a good and succinct description of two major AI control tools for humanoid robotics: Vision-language-action (VLA) models for practical tasks; Motion retargeting for walking in a way that looks cool - YouTube

Seen on social media somewhere: “my 75yo mother told me last night that she’s planning to delete Instagram and Facebook because she’s sick of not being able to tell what’s AI and what’s not, and I genuinely don’t think tech companies have reckoned with this kind of move as an actual possibility”

Follow the latest AI headlines via SkynetAndChill.com on Bluesky