SkynetAndChill by druce.ai
Posts
AI Reading for Tuesday December 30

AI Reading for Tuesday December 30

Manu$ and In$ilico ca$h in

Druce Vertes
December 30, 2025

Meta will acquire Chinese-founded AI company Manus, which says it has $100 million ARR eight months after launch, serves millions daily, its staff will join Meta, and Manus will sever ties to Chinese investors while deal terms remain undisclosed. - The Register

Price said to be $2-3b, would assume higher end of range because if you can’t get a $1 billion premium from Zuck what are you even doing? Will sever links to China, operations to remain in Singapore for now.

Meta will integrate Manus's autonomous general-purpose agent into its products, with Manus's team joining Meta while Manus continues selling its standalone service; Manus's agent has processed over 147 trillion tokens and created over 80 million virtual computers. - Meta for Business

Manus, which processed over 147 trillion tokens and powered more than 80 million virtual computers, is joining Meta while remaining based in Singapore and aims to scale to millions of businesses and billions of Meta users, CEO Xiao Hong said. - Manus

OpenAI faces a make-or-break 2026 as CEO Sam Altman expands beyond chatbots into custom chips, e-commerce, consulting and a consumer device, seeking new revenue while risking overextension and heavy capital costs. - The Economist

Big story of 2025 was them losing their lead. The price-performance envelope is getting very square, high degree of commoditization and not much bang for the buck past Gemini 3 Flash. Artificial Analysis has Nemotron Nemo Nano and and MiMo in interesting spots.

Geoffrey Fowler wrote in a Washington Post column that after testing ChatGPT, Claude, Gemini and other chatbots he concluded ChatGPT is overrated, advising users to pick models by task rather than brand and foreseeing continued fragmentation. - The Washington Post

SoftBank completed its $40 billion investment commitment in OpenAI, sending a final $22–22.5 billion tranche last week and raising its stake above 10% at a reported $260 billion pre-money valuation, sources told CNBC’s David Faber. - CNBC

Wanted: Director of Doomsday Prepping: Sam Altman posted a search for an OpenAI Head of Preparedness offering a $555,000 base salary plus equity to secure systems, lead preparedness for frontier AI capabilities, and address rising risks after multiple safety-team departures. - The Register

Founder/beast mode: Microsoft CEO Satya Nadella has overhauled the company's leadership to pursue an AI strategy beyond its frenemy OpenAI partnership, Alphabet, Amazon and start-ups. - Financial Times

SoftBank agreed to acquire DigitalBridge Group for $4.04 billion in cash at $16 per share (about 16% premium), buying a firm with roughly $108 billion AUM and CEO Marc Ganzi to remain; deal expected to close in H2 2026. - SiliconANGLE

seems low, stock peaked at twice the price, either prospects appear diminished or a bit of sweetheart deal shenanigans

Surging AI-driven data center demand could drive consumption toward ~2,200 TWh and cause mined uranium to meet under 75% of reactor needs, prompting investor forecasts of $100-135 per pound of uranium by 2026. - OilPrice.com

VC: "The 'build it and they will come' strategy is a trap. If you are a hyperscaler, you will own your own data centers. We foresee a significant financing crisis in 2027–2028 for speculative landlords." - Axios

In 2025, major AI firms raised tens of billions—OpenAI reportedly $40 billion at a $300 billion valuation, Anthropic $16.5 billion (valuation $183 billion), and xAI about $10 billion—while companies pledged roughly $1.3 trillion in infrastructure amid lawsuits and regulatory scrutiny. - TechCrunch

The year everyone invested in everyone: In 2025 major AI deals included Disney's $1 billion OpenAI investment, Google's ~$32 billion Wiz buy, Meta's $14.3 billion for 49% of Scale AI, a U.S. $8.9 billion 9.9% Intel stake, and Stargate's $500 billion buildout. - Business Insider

You’re more hedged but everybody rises or falls together.

Wait, you guys have use cases? Reporters discuss how real companies are using AI. - The Wall Street Journal

The AI Scorecard reports the United States built an AI lead but might lose it to China. - The Wall Street Journal

China channels Elizabeth Warren, drafts the world's strongest AI consumer protections, with immediate human intervention and guardian notification when AI mentions suicide, mandatory guardian contact information for minors and the elderly, and bans on violent or emotionally manipulative content. - Ars Technica

Libertarian and authoritarian right both hate AI bill - Reason.com

passes for bipartisanship

Hot girl propaganda with beauty filters: TikTok account 'Prawilne_Polki' used AI-generated videos with synthetic faces and desynchronized audio to promote Polexit to 15–25‑year-olds, garnering about 200,000 impressions and 20,000 likes before being deleted after rebranding on 13 December. - Euronews

Social science to science fiction: Dartmouth researcher Sean J. Westwood built a two-layer autonomous agent that can mimic human survey responses, risking large-scale contamination of digital surveys and prompting his warning that current safeguards are insufficient. - Fast Company

Trump Tower Gaza? Where does AI-driven "political simulation" become planning of genocide and speedrun from war game to war crime? - Global Research

Prior AuthGPT: Medicare will launch a Jan. 1 to Dec. 31 pilot in six states, including New Jersey, using private AI/ML vendors for prior authorization of certain higher-risk services, tying vendor payments to savings and requiring human clinician review of denials. - nj

WebMD now MDGPT: Patients are increasingly consulting ChatGPT and other AI chatbots before appointments as hundreds of new oncology studies appear daily, prompting proposals for ACGME-mandated AI transparency and standards by 2026, board-exam AI literacy within two years, and Dartmouth Geisel's AI curriculum. - STAT

Better living through AI chemistry: Insilico, an AI drug discovery startup, rose 45% in its Hong Kong trading debut after raising $293 million in an IPO and reported more than $85 million in revenue in 2024. - Bloomberg

Four new books examine LLM chatbots and therapy apps used to address a global mental-health crisis affecting over a billion people, reporting mixed results and privacy, monetization, and safety concerns, with OpenAI saying 0.15% of users share explicit suicidal indicators. - MIT Technology Review

Investors rotating into understandable AI "pick-and-shovel" names supplying hyperscaler data-center buildou ts, over AI magic with magic multiples. Data-storage and infrastructure stocks dominated the S&P 500 in 2025, with Sandisk up about 580% as the index's best performer, Western Digital second - Yahoo

Strategist Hardika Singh said the Federal Reserve increasingly sees AI-driven productivity raising output without stoking inflation, lifting 2026 growth expectations and corporate profits while concentrating risk among high-beta AI beneficiaries like Micron. - Yahoo

Fed researchers can see whatever they want, but a new political hack at the Fed will drop rates regardless of what they see for AI, the economy, or anything else.

Trump: "We have a fool at the Federal Reserve. I mean, Biden reappointed him. It's too bad. You would have thought he wouldn't have done that."

(Jerome Powell was appointed by Trump over Yeltsin, who ironically might have been a bit more dovish but he thought less beholden.)

The A.I. boom produced a new crop of paper billionaires as valuations surged—e.g., Scale AI founders after Meta's $14.3 billion investment and Cursor co‑founders after a $27 billion valuation. - The New York Times

In 2025 AI shifted from chatbots to tool-using autonomous agents driven by Anthropic's Model Context Protocol and Google's April 2025 Agent2Agent (donated to the Linux Foundation), prompting vendor releases, agentic browsers and security and governance challenges. - The Conversation

Three long-term memory types—episodic, semantic and procedural—are needed for autonomous AI agents; short-term context windows are insufficient, prompting hybrid memory architectures, LLM–vector DB convergence, and increased demand for memory-design engineering. - MachineLearningMastery.com

Quantum computing inched toward actual relevance, while focus was on artificial intelligence. - Bloomberg.com

Imagine what Linus Torvalds could have done in 1991 with Claude Code. Now imagine that there are geniuses out there doing it.

Some kid out there is making a mashup of BeOS and Plan9 in Rust or OCamL, with native AI and virtualization and zfs and whatnot, and porting GNU, all the standard libraries and all the functionality of Linux with AI. Or maybe JavaScriptOS LOL.