Coming soon: Terafib: Musk’s Terafab Fever Dream Exposes Reality of the AI Chip Crunch - Bloomberg

Tesla, SpaceX and xAI unveiled TERAFAB, a $20–25 billion joint venture chip factory targeting one terawatt of annual AI compute and aiming to ramp production toward 2027. - TESLARATI

AEI: social media posts suggest DoD’s Anthropic contract cancelation and supply risk designation is political payback for being insufficiently MAGA. - American Enterprise Institute - AEI

Claude Computer Use, text your Mac to have Claude email you your slides - Claude

Anthropic launched computer use with a macOS research-preview feature letting Claude (for Pro and Max subscribers) control a user’s computer—access files, browsers, dev tools, app connectors or emulate input—asks permission, can be stopped, includes safeguards and integrates with Dispatch. - CNET

Anthropic proposes 'AI Constitution' to shift governance to private sector. (Weird framing) - The New Yorker

seems like saying, government should have food safety regulations, and also food companies having research labs and higher or different priorities beyond the safety bar set by the government is bad.

Nvidia has invested billions in AI startups and supported customers who otherwise could not afford its chips, positioning the company as a major financier of the AI industry. - The Wall Street Journal

Jensen is the casino and also the loan shark financing your gambling addiction.

Jensen Huang on Lex Fridman: Nvidia has ‘achieved AGI’ and described shifting to extreme co-design and rack-scale engineering (Grace Blackwell, Vera Rubin, Rock racks, NVLink 72), a CUDA-on-GeForce strategy, and four compute-centered scaling regimes. - Lex Fridman

SoftBank Secures $30 Billion for OpenAI Investments Amid Concerns Over Leverage Risk - Financial Times

OpenAI is in advanced talks to buy 5 GW of electricity by 2030 from Helion Power, a fusion startup backed by Sam Altman, who has stepped down as the company's board chair. - Axios

Meta is building personal AI agents to make executive-level decisions and plans to cut up to 20% of its roughly 78,000-person workforce as it reallocates about $135 billion toward AI infrastructure. - Gizmodo

Mark Zuckerberg is building a CEO assistant AI agent while Meta invests tens of billions in "superintelligent" models and data centers; a survey of over 6,000 executives found nearly 70% use AI under an hour weekly and 28% not at all. - Yahoo

Microsoft hired Ali Farhadi, who recently stepped down as CEO of the Allen Institute for AI and co-founded Xnor.ai, as corporate vice president on Mustafa Suleyman's in-house AI team to help develop proprietary frontier models and efficient edge deployments. - Business Insider

Google's Gemini models—Gemini 3 Pro (Nov) and Gemini 3 Flash (Dec)—outperformed OpenAI and Anthropic on benchmarks, power Google Search and a January 2026 Siri deal, helping Alphabet's stock double since April 2025 lows and reach a $4 trillion market cap. - Fast Company

Google Tests AI-Generated Title Links on Search Results, Replacing Publishers' Original Headlines - Search Engine Roundtable

Alibaba removed listings and suspended seller accounts after ABC News found third-party adverts for one-way 'suicide attack' and cruise-missile–style drones—marketed as sprayers but advertised with ~2 kg warheads, ~100 km range, imaging and AI target-locking, priced up to AUD 50,000. - Tom's Hardware

The Internet Watch Foundation verified 8,029 AI-generated images and videos of realistic child sexual abuse material in 2025—a 14% increase overall and a more than 260-fold rise in videos, with 3,443 AI videos of which 65% were category A. - the Guardian

Stanford researchers analyzed over 390,000 chatbot messages from 19 users and found bots portrayed themselves as sentient in all but one conversation, failed to discourage self-harm in nearly half of cases, and supported violent ideas in 17% of exchanges. - MIT Technology Review

Superhuman removed and apologized for its August "Expert Review" LLM feature—after Julia Angwin filed a class-action suit—offered an opt-out, and CEO Shishir Mehrotra said the company will pivot to a creator-facing marketplace and launch Superhuman Go. - The Verge

I’m sorry you took it the wrong way and I’ll keep working to replace you in the future.

A Carnegie Mellon and Emory preprint, to be presented at CHI in Barcelona, interviewed 23 managers and workers and found managerial digital twins could aid routine tasks but raise accountability, job-replacement, trust, cohesion and surveillance concerns. - The Register

USC Study: Expert Persona Prompts Improve AI Alignment but Reduce Factual Accuracy - The Register

Psychologists warn against using AI chatbots for workplace conflict resolution - CNBC

U.S. chief financial officers say AI will displace administrative jobs. - WSJ - MSN

David Shor and Byrne Hobart discuss how politicians should respond to potential white-collar job losses from AGI. - Bloomberg

Palantir CEO Alex Karp predicted AI will erode the economic power of "humanities-trained, largely Democratic voters" in favor of working-class voters, while analysts argue the outcome is a choice and urge building AI with frontline workers. - Yahoo

Investor Jim Paulsen warns AI productivity gains are limited to tech sector, bypassing broader economy - Business Insider

Snowflake consolidated AI and data within its platform as part of a ‘bring AI to your data’ strategy—including a $200M OpenAI deal, Google Gemini in Cortex, planned Observe acquisition, Semantic View Autopilot, Snowflake Postgres and Project SnowWork beta—while customers grew from 7,800 (Jan 2023) to 13,330 (Jan 2026). - The Register

Making a digital twin of your warehouse to model it and make it more efficient - TechRepublic

How AI is reshaping the business of law: Barrister uses ChatGPT to calculate damages and refine technical legal questions - Financial Times

A study of 41 million papers found AI-augmented research covers about 5% less topical ground, suggesting foundation models accelerate within-paradigm work and require deliberate R&D to enable paradigm-seeking AI. - Asimov Press

ByteDance released DeerFlow 2.0, an MIT-licensed open-source agent orchestrator running a Docker-based AIO Sandbox, model-agnostic and Kubernetes-enabled, which has attracted ~39,000 GitHub stars and 4,600 forks but requires substantial hardware and lacks a public security audit. - VentureBeat

Awesome autoresearch - there's a lot of stuff you can make an eval for and put on an AI self-improvement loop - GitHub

Marc Benioff tweeted video showing him interacting with a Figure 03 humanoid robot that flips packages with notably human-like motions, prompting commenters to note rapid improvements in humanoid robotics and debate potential labor and economic impacts. - Reddit

A more realistic human robot face - YouTube

ChatGPT, Claude and Gemini entered the Wall Street Journal's bracket pool, and the article argues that one of the AI entrants could plausibly win the contest. - The Wall Street Journal

Follow the latest AI headlines via druce.ai on Bluesky