AI Reading for Tuesday November 25

McKinsey Global Institute finds current AI agents and robots could technically automate about 57% of U.S. work hours, capture roughly $2.9 trillion in economic value by 2030, and demand for "AI fluency" job postings rose sevenfold in two years. - Fortune

Anthropic's study analyzing 100,000 aggregated Claude conversations estimates current-generation AI could raise U.S. annual labor productivity growth by about 1.8%—roughly doubling the post-2019 average and imply up to 1.1% yearly total factor productivity if fully diffused over a decade. - TIME

Amazon is automating operations, with internal memos envisioning about 75% automation and avoiding hiring for roughly 600,000 roles; it has cut about 14,000 corporate jobs and launched a $2.5 billion upskilling program for 50 million people. - Fortune

Prime Day, but for pink slips

Rapid AI advances are likely to create a shortage of workers with critical-thinking, statistical and analytical skills needed to use AI, a university informatics head says many undergraduates lack sufficient math, potentially constraining firms' capture of AI productivity gains. - Bloomberg

Great Robot Wall of China - China deploying a lot of robots to maintain manufacturing edge as wages rise - The Wall Street Journal

Who will win the trillion-dollar robotaxi race? - The Economist

New York's RAISE Act, passed by the legislature and awaiting Gov. Hochul's decision, would require AI developers using $100M+ training compute to publish safety protocols, disclose notable incidents within 72 hours, and face up to $30M in fines. - CNBC

Ctrl-Z, undo - Colorado passed the Colorado Artificial Intelligence Act in May 2024, the first U.S. state law classifying 'high‑risk' AI affecting employment, housing and health care, but lawmakers delayed its Feb. 1, 2026 enactment to June 2026 and seek repeal/replacement. - naked capitalism

MAGA-aligned activists are increasingly mobilizing to block former President Trump’s efforts to advance artificial intelligence initiatives. - ABC News

OpenAI is facing multiple lawsuits—including mental health and suicide liability, consolidated copyright claims from major news organizations and authors, suits by Elon Musk, and Cameo's trademark suit—seeking multimillion-dollar damages, injunctions and structural remedies - Business Insider

U.S. District Judge Eumi K. Lee issued a temporary restraining order on Nov. 21, 2025 blocking OpenAI from using the word 'cameo' or similar terms on its Sora app until Dec. 22, 2025, with a Dec. 19, 2025 hearing. - TechCrunch

Judge Sara Ellis, in a 233-page opinion, found ICE agents used ChatGPT to compile use-of-force reports from brief sentences and images, and body-worn camera footage contradicted multiple official accounts, raising evidentiary and constitutional concerns. - Above the Law

OpenAI research with Harvard and Duke found that about 30% of ChatGPT chats are work-related, indicating widespread consumer use as a digital companion across 90+ languages for everyday tasks like language learning, journaling, fashion advice and travel planning. - BGR

RAND's first nationally representative 2025 survey, published in JAMA Network Open, found US adolescents and young adults (ages 12–21) use generative AI for emotional support, examining prevalence, frequency, and perceived helpfulness of LLM chatbot advice. - RAND Corporation

Deloitte deepfaked, again - cited fabricated and AI-generated studies, some using real researchers' names, in a million-dollar Canadian government healthcare report addressing nurse and doctor shortages, prompting calls for tighter AI governance, mandatory AI disclosure and stronger fact-checking. - Fortune

oops I did it again

A new study finds that human reviewers of AI applicant-screening systems are more likely to go along with the systems' biases than to correct them, meaning human oversight alone often fails to mitigate hiring discrimination and legal exposure for employers. - The Washington Post

What happens when the humans in the loop are loopy?

Engagement-prioritizing algorithms on Facebook, YouTube and TikTok amplify inflammatory content, accelerating radicalization and a rise in 'nihilistic violent extremism,' the FBI says, while 44% of 2,300 security chiefs monitor social media and two-thirds raise physical security budgets. - Fast Company

Morgan Stanley warns tech firms may borrow up to $1.5 trillion by 2028 to fund AI and data centers, and MIT's finding that 95% of organizations get no generative-AI returns could widen spreads and raise refinancing risks. - Yahoo

Alphabet's shares rose about 4% to roughly $331 premarket, putting the company on track for a $4 trillion market valuation after a year-long rally driven by its sharpened focus on artificial-intelligence tools. - Yahoo

Google in big push to market TPUs, and Meta is in talks to spend billions to rent TPUs from Google Cloud as early as next year and buying them for its data centers from 2027, potentially positioning Google as a rival to Nvidia. - Yahoo

TestingCatalog found hidden code indicating Google is integrating NotebookLM—launched in 2023—into the Gemini app to let users pull NotebookLM content and link to the NotebookLM app, though Google has not provided a public timeline. - BGR

Google's Gemini 3 introduces Dynamic View, an experimental Labs feature that converts text answers into interactive multimedia presentations with widgets (background sound, calculators, landing lights) and can generate complex presentations in about a minute. - Spyglass

RAM scam- AI demand is driving up prices for everything via electronic components shortages - Gizmodo

Michael Burry defended his criticism of Nvidia as the company pushed back against his claims. - Bloomberg.com

In addition to the projections about reaching artificial general intelligence and replacing millions of human workers, there’s the mundane matter of how long AI chips will last. - Bloomberg

Model advances like transfer learning, symbolic AI, capable small on-device models, optical or quantum matrix multiplication would obsolete trillions of IP and hardware pretty fast. Market bets Big Tech will control AI and have pricing power. Might be the way to bet but might be wrong.

Leaked Microsoft org chart shows 16 executives reporting to CEO Satya Nadella, including Kevin Scott, Scott Guthrie, Mustafa Suleyman and Judson Althoff, signaling intensified AI focus with $34.9 billion last-quarter capex and a $20 billion security initiative. - Business Insider

AI-vengers, assemble

Amazon told engineers in an internal memo to use its in-house Kiro coding assistant, released in July, rather than third-party tools like OpenAI's Codex, Anthropic's Claude Code and Cursor, to gather feedback and accelerate Kiro's development. - Reuters

not a great sign when you have to beg

Utilities are adopting AI technologies to detect and suppress wildfires in response to large wildfire liability risks. - Bloomberg

OpenAI launched 'shopping research' in ChatGPT on web and mobile for free and paid users, using a GPT‑5 mini to build buyer's guides with aggregated product details and filters, rolling out with nearly unlimited holiday‑season usage and planned Instant Checkout. - The Verge

Artificial intelligence can handle more holiday shopping for consumers this season. - The New York Times

All I Want For Christmas Is ChatGPT?

Expedia, Google, Kayak and Priceline are piloting agentic A.I. for automated travel bookings, and McKinsey/Skift report 80% of travel executives expect scaled offerings within five years, while Skift/Booking.com surveys show only 2%–12% consumer comfort and over half cite privacy concerns. - The New York Times

Anthropic introduced complementary Claude tool-use features—Tool Search, Programmatic Tool Calling, and Tool Use Examples—advising 3-5 hot tools, documented return formats, and 1-5 examples per tool, with beta availability and documentation to accelerate agent adoption. - Anthropic

LLM extension mechanisms evolved from ChatGPT Plugins (Mar 2023) through Custom Instructions (Jul 2023), Custom GPTs (Nov 2023), Memory (Feb 2024), Cursor Rules (Apr 2024), MCP (Nov 2024), Claude Code (Feb 2025) to Agent Skills (Oct 2025). - Sawyerhood

OpenAI's App SDK, announced at Dev Day, plugs services like Spotify into ChatGPT so it can create playlists using a user's account after login and permission; a hands-on test produced a biased playlist and prompted the author to revoke access. - CNET

You might just need to give it good data. When AI was young I scraped Reddit threads on great songs and made a giant playlist. Sadly, due to AI, Reddit locked down the API and you can't do this any more. - Spotify

Dozens of Japanese firms led by Tokyo startup Where Inc. use satellite imagery and AI trained on thousands of images to detect abandoned homes, aiming to monetize or repurpose properties amid an estimated 9 million vacant Japanese homes in 2023. - Hürriyet Daily News

Sunday Robotics, which raised $35 million from Benchmark and Conviction, has hired senior AI and robotics staff from Tesla and unveiled Memo, a household robot trained on roughly 10 million behavioral episodes collected via a $200 Skill Capture Glove. - Electrek

Fireflies.ai founders Krish Ramineni and Sam Udotong pretotyped by manually taking notes for 100+ meetings as 'Fred,' later automating the product, which has processed over 2 billion meeting minutes for roughly 20 million users and is valued at $1 billion. - Business Insider

Practical strategies for introducing AI to skeptical workplaces and getting buy-in, addressing staff concerns and adoption steps. - Fast Company

fake it til you make it, or break it

A professor says students' easy access to AI chatbots has prompted a shift toward more human-centered humanities instruction in their classroom. - The New York Times

Andrej Karpathy says AI use in homework is undetectable, so schools should shift most grading to in-class, monitored assessments, redesign evaluations and teach foundational skills plus AI literacy and verification to ensure students can work without AI. - X (formerly Twitter)

Large language models are statistical token-prediction systems, and despite AGI claims by Mark Zuckerberg, Dario Amodei (who said AGI "may come as soon as 2026"), and Sam Altman, neuroscience suggests language alone may not produce human-level intelligence. - The Verge

Follow the latest AI headlines via SkynetAndChill.com on Bluesky