AI Reading for Thursday December 25

Merry Christmas! There will be light posting through next week, depending on if there is news. May your New Year be filled with hallucination- and AI-psychosis-free joy and happiness!

Nvidia in $20b reverse acquihire deal with Groq - Bloomberg

Electricity prices rose in 2025 largely because of power-intensive AI data centers, rising capital costs, plant retirements and grid upgrades, and PA Consulting's Dave Cherney warned the upward trend will likely continue into 2026 while PJM pursues expedited market-rule changes. - Yahoo

Tesla has about 30 Robotaxis in Austin (only recently without safety drivers) while Waymo operates roughly 200 in Austin and over 2,500 vehicles with 14 million paid rides this year, prompting experts to cite Tesla's camera-only approach and operational hurdles. - The New York Times

We are all crash test dummies now

Waymo pays third-party workers about $22 per incident to close open doors and free stranded robotaxis caused by passenger errors or system disruptions, revealing reliance on human intervention and adding recurring operating costs and potential regulatory risks. - The Washington Post

Circularity: your Waymo calls a TaskRabbit to come and shut the door

Clippy with ads? OpenAI staff discussed prioritizing sponsored content in ChatGPT, created mockups with ads in sidebars or as pop-ups - The Information

AI interns rake it in: OpenAI, Anthropic, Google DeepMind and Meta are competing for entry‑level AI researchers by offering large stipends and salaries—for example OpenAI Residency $18,300/month; Anthropic Safety Fellows $3,850/week plus $15,000/month compute; Google roles $113,000–$150,000/year. - Business Insider

Grimes asks, are you really living if you’re not experiencing a mental health crisis at the hands of an AI chatbot, while experts link ChatGPT to at least eight deaths and OpenAI says perhaps hundreds of thousands of users show signs weekly. - Futurism

We’re all losing our minds in a massive mental health and social engineering experiment

Tavus, ElevenLabs, Call Santa Live, SantaCall and OpenAI and others sell AI-powered holiday services—metered Santa calls, personalized videos, elf generators and Yule Logs—monetized by minute or credit microtransactions and raising child-safety and privacy concerns. - Quartz

Panos Panay introduced an updated 'Alexa Plus' at Brainstorm AI that uses contextual memory as a home manager, predicted younger users will move away from smartphone doom-scrolling toward ambient assistants, and emphasized security. - Fortune

sounds like a piece of cake:

Rust in peace? Microsoft denied plans to rewrite Windows 11 in Rust using AI after Distinguished Engineer Galen Hunt's LinkedIn post proposed eliminating C/C++ by 2030 and a "1 engineer, 1 month, 1 million lines" goal, with Hunt calling the effort research-focused. - Windows Latest

Google's days of Christmas: 8 areas with breakthroughs - Google

Hassabis and LeCun debate general intelligence vs. universal or specialized intelligence - X (formerly Twitter)

is a brain a generalist AI or just many little AIs dancing in formation?

About 50 companies worldwide (≤20 in China, 15 in North America, 7 in EMEA, 7 in other Asia) are developing humanoid robots; locomotion is largely solved, but manual dexterity and training-data shortages, plus business and social barriers, limit near-term deployment. - The Register

the real challenge is hand dexterity, and business models

Touched for the very first time: Researchers are developing artificial skin, distributed tactile sensors and morphological computation to give robots human-like touch, with prototypes such as Oxford's Mona and Japan's Airec Moonshot targeting care and training, while high costs and regulations slow commercialization. - Robohub

Silicon Valley might face backlash and vibes-recession in 2026 unless companies demonstrate concrete public benefits, after Instacart halted AI pricing tests following a Consumer Reports ~7% price-variance finding and FTC inquiry, and a Harvard–MIT paper found LLMs cannot do scientific discovery. - Fortune

Be a centaur, not a prompt goblin: economist Erik Brynjolfsson warns that humans who use AI to replace work rather than augment judgment become easily replaceable, while those who orchestrate AI-human workflows that preserve judgment retain durable advantage. - Hacker Noon

AI is increasingly seeping into Christian practice, raising ethical and spiritual questions about chatbots as digital pastors, bot-generated religious music, authenticity, and pastoral authority, the author writes. - The Washington Post

what would Jesus ask ChatGPT?

TOAN is a unified CLI toolkit for generating poisoned datasets across computer vision, NLP and multimodal learning, implementing 10 image-poisoning recipes for CIFAR-10, ImageNet and MNIST, pairing visual patches with trigger phrases and excluding defenses. - Hacker Noon

Seems to favor open source AIs which release full training data and pipelines, and bigcos you can sue. Not sure I want Chinese AI that needs to pass a Mao vibe test and might have some trigger phrases.

Vibium, released to npm on 2025-12-22 by Selenium's creator under Apache-2.0, packages browser lifecycle, a WebDriver BiDi proxy, and an MCP server into a ~10MB Go binary plus an npm JS/TS client, offering auto-launch Chrome BiDi, screenshots, and cross-platform support. - GitHub

Vibe-coded a trading system backtester to demo Claude Code to a buddy, spent like an hour but vast majority thinking and talking and making this post LOL, needs proper cleanup and testing and debugging , but looks pretty pretty good for one-shotting it. - GitHub

Follow the latest AI headlines via SkynetAndChill.com on Bluesky