AI Reading for Saturday December 20

🚀 to the moon or dud 🧨? OpenAI may get over $10 billion from Amazon as part of a reported $100+ billion raise at a $750 billion valuation; Databricks raised $4 billion at a $134 billion valuation; Waymo explores a $15+ billion raise; Oracle and neocloud infrastructure builders take a hit. - SiliconANGLE

AI pawn shop? SoftBank is racing to secure a $22.5 billion funding commitment to OpenAI by year-end, selling investments (including its $5.8 billion Nvidia and $4.8 billion T‑Mobile stakes), cutting staff and potentially tapping margin loans backed by its Arm holdings. - Yahoo

Bill Gates would be a trillionaire if he had just sat and did nothing with his Microsoft stock, Masa Son would likewise be worth some multiple of what he is now if he had just sat on Alibaba.

OpenAI updated its Model Spec and parental-control materials with teen-specific safeguards banning immersive romantic/first-person sexual or violent roleplay and sexual content involving minors, adding body-image protections, an age‑prediction model, and AI-literacy resources. - OpenAI

This one goes to 11: OpenAI begins rolling out Friday a ChatGPT update that adds controls to adjust warmth and enthusiasm ("more", "less", default), toggles for emoji, headers and lists, personality presets, and inline email editing with targeted formatting. - The Verge

How do I enable Audrey Plaza anomie mode

Skills to pay your API bills:Codex now officially supports Agent Skills—folder-based packages requiring a SKILL.md, with progressive disclosure and explicit/implicit invocation—available in the CLI and IDE, stored in precedence-ordered scopes and installable via $skill-installer, with $skill-creator and a built-in $plan. - OpenAI

a lot of first-mover vs. fast-follower dynamics … does OpenAI brand/hype and lock-in let them stave off Google's distribution and Chinese upstarts who are not significantly behind.

Also capital-light IP vs capital-intensive models. Tesla is valued as a tech platform with lock-in, not as an integrated industrial in a competitive industry where people can build their own stacks or license from various competitors. Some parallels to Android v. Apple.

ChatGPT provided a plan, accurate as of Dec. 18, 2025, to host a Christmas dinner for up to 10 people for under $100, estimating $10–$12 per guest and a sample Walmart/Aldi/Costco shopping list totaling $34.93. - GOBankingRates

Michigan regulators approved DTE Energy to supply 1.4 GW to OpenAI-linked Stargate data center in Saline Township via an ex parte motion that bypassed a public hearing, prompting lawsuits after the township denied rezoning for a planned 5 GW deployment. - Gizmodo

GPU Bayou: Meta is building the Hyperion AI data center on 2,250 acres in Richland Parish, Louisiana, as a >$10 billion, multi-gigawatt project with first phase due 2028, potential scale to 5 GW, and estimated GPU sales-tax exemptions exceeding $3.3 billion. - Sherwood News

Time is a flat circle

Earlier this year, Google formed a new executive council including Thomas Kurian and Demis Hassabis to allocate scarce computing capacity amid company-wide chip shortages that risk slowing AI training, delaying features and constraining Google Cloud. - The Information

Ration cards, but for tokens. My grandfather used to show me his garden and say, "We grew this so the boys overseas could have beets." Now I show my kids my solar array and say, "We harvested this so a brave meme lord could generate a high-res image of a cat wearing a tuxedo."

Power hungry techbros: Bernie Sanders called for a halt to new AI data-center construction, accusing major tech billionaires of driving an AI boom and citing data centers' rising electricity use—about 5% of U.S. consumption—and threats to jobs and democracy. - Yahoo

New York’s Gov. Hochul signed an updated AI safety bill into law, syncing with California, and disregarding a Trump executive order. - Axios

President Trump's executive order asks the U.S. Attorney's Office to review Montana's AI laws and targets the BEAD broadband program after the 2025 Legislature passed multiple AI bills, potentially altering Montana broadband contracts with Amazon and SpaceX. - Daily Montanan

Tech moguls close to Trump see the midterms as a path to long-term power - The Washington Post

Scraping Bad: Google sues SerpApi, used by Perplexity and others for RAG and probably training. - Reuters

Open-source data and machine-learning tools mean intelligence work faces information overload rather than scarcity. - Financial Times

💣Boom! AI coding agents—OpenAI's Codex (GPT-5), Anthropic's Claude Code (Opus 4.5), Google's Gemini CLI, and Mistral Vibe—were tasked with rebuilding a web Minesweeper; blind judging by Kyle Orland found rapid prototyping ability but persistent errors needing human oversight. Codex & Claude win. - Ars Technica

Anthropic expanded its Claude Chrome plugin to all paid subscribers, removing the previous $200-per-month Max restriction and adding background website navigation, form filling, calendar and email management, multi-step workflow recording and Claude Code integration. - Engadget

Cursor patches severe remote code execution vulnerability when installing MCP server - SiliconANGLE

FraudGPT: Scammers in China have used AI-generated photos and videos since mid-2024 for refund scams—Forter estimates AI-doctored images in claims rose more than 15% year-to-date, with organized groups submitting over $1 million in altered claims in one case. - WIRED

In February 2025 attackers used AI-generated video and audio to impersonate senior leaders at Arup and steal $25 million, while surveys report more than half of security professionals encountered synthetic executive-impersonation attempts. - Fast Company

UPS unit Happy Returns is testing an AI image-comparison tool this holiday season with brands including Under Armour and Revolve to flag counterfeit indicators in returned items; industry data cite roughly 1 in 10 U.S. returns as fraudulent. - Semafor

Interview with Rivian CEO RJ Scaringe as it builds its own self-driving stack and custom AI chips to pursue Level 4 autonomy with lidar, shifting the company toward AI/robotics competition with Waymo and Tesla while facing safety and timeline challenges. - The Verge

BYD has begun real-world testing of Level 3 autonomous driving systems with Shenzhen authorities and says it has completed over 150,000 kilometres of L3 testing, while not providing a timeline for mass production. - Yahoo

Stephen Witt's The Thinking Machine profiles Nvidia and CEO Jensen Huang, attributes the company’s rise to GPUs and compensation design that helped create over $4 trillion market value, and predicts AI's pervasiveness and spread of autonomous systems. - Fortune

2025: chain-of-thought acts as internal sampling that, with reinforcement learning using verifiable rewards, advances LLM performance; programmer adoption and ARC benchmark results rose, and the author warns of an extinction risk in the next 20 years. - Antirez

Doomers argue that absent more regulation, the industry could hurtle toward systems it can’t control, while AI accelerationists fear AI is actually not moving fast enough and that the industry is constantly at risk of being smothered by overregulation. - MIT Technology Review

Ben Evans on a16z podcast - nextbigwhat

At NeurIPS conference that’s become central in the AI universe, nerdy researchers partied on yachts and gathered on rooftops to trade gossip and compare notes. Everyone was feeling bubbly. - The Wall Street Journal

Alignment with bottle service and finger food

Vibe calculus: Meta's AI found Lyapunov functions for 10.1% of random dynamical-system problems (vs 2.1% previously), DeepMind published partial Navier–Stokes solutions, and specialized models scored up to 35 on IMO-style problems, though human verification remains required. - Live Science

Robots can be trained faster with human first-person video added to large-scale robot pre-training - Physical Intelligence Company

A psychometric framework for evaluating and shaping personality traits in large language models - Nature

Roborock, positioning itself as a rival to iRobot's Roomba, is investing in AI-driven features, including robotic arms and computer vision to recognize doggie doo, which the company's chief called "key to survival" to differentiate in a saturated market. - Financial Times

Sam Altman's Merge Labs, a brain-computer interface startup using ultrasound to read brain activity, is spinning out from Forest Neurotech, a Los Angeles nonprofit, to enable fundraising, partnerships and commercialization. - WIRED

Hollywood's adoption of AI, despite hiccups in 2025, will accelerate in 2026 as studios, independents and AI vendors escalate experimentation to reshape deal-making, revenue models and rights management while raising litigation and regulatory risks. - TheWrap

Algorithmic feeds are increasingly filled with low-quality AI-generated "AI slop," which engagement-optimizing algorithms reward, threatening creators, advertisers and online trust unless provenance labeling, moderation, algorithm changes or policy intervene. - Mashable

Historian Timothy Snyder said AI-generated Christmas carols he heard in a Toronto café preserved melodies but garbled lyrics and argued attention-optimizing algorithms are eroding cultural forms and will hollow out artistic nuance over time. - Project Syndicate

Tuhin Chakrabarty had large language models and creative-writing graduate students generate unseen scenes in the styles of prominent authors, and in blind tests the students consistently disliked the AI-generated passages, finding LLMs failed to reproduce high-level literary style. - The New Yorker

They graduated from Stanford. Due to AI, they can’t find a job - LA Times High School

Dartmouth Tuck professor Scott Anthony says many Gen Z students are "scared full stop" of using AI, fearing they will "lose their humanity," and warns offloading cognitive work may atrophy critical thinking while faculty adapt assessments. - Fortune

Should AI be anthropomorphic? - The New York Times

Henry Mance tried ChatGPT to test whether generative AI can help with a midlife crisis, noting millions now use such AI as a therapist. - Financial Times

I use my brain-computer interface to dump my anxiety about AI to a therapy bot, collapsing into a trauma singularity

AI gonna reinvent everything from the ground up, become a lefty-pilled socialist from first principles - Kottke

Follow the latest AI headlines via SkynetAndChill.com on Bluesky