AI Reading for Friday November 14

OpenAI's GPT-5.1 available in API, adds adaptive reasoning, a 'no reasoning' fast-answer mode, 24-hour prompt caching, Codex 5.1 and new tools, reducing latency and token use to cut API costs and speed multi-step coding, agent workflows and embedded app performance. - OpenAI

guess I know what I’m working on this weekend

OpenAI is piloting group chats in ChatGPT in Japan, New Zealand, South Korea and Taiwan that let up to 20 logged-in Free, Go, Plus and Pro users collaborate using GPT-5.1 Auto with image/file upload, generation and no personal memories. - OpenAI Help Center

Battle brewing over the future of agents and online shopping - CNN

Major retailers are battling chatbot developers over access to user behavioral and tracking data, with consent controls limited to GDPR countries and some US states, making fully autonomous consumer shopping agents unlikely until privacy, tracking and interoperability issues are resolved. - WIRED

Shopping Cart Blanche? Google introduced AI shopping features using its Shopping Graph of over 50 billion listings, including automated store calls and an agentic checkout that can make purchases via Google Pay with user confirmation in a U.S.-first rollout. - Mashable

if you’re Google and you have Chrome and Search data, you already know everything and don’t need permission

Sundar Pichai acted as a "wartime CEO" after ChatGPT's launch, helping make Google an AI leader by empowering Demis Hassabis. - Bloomberg

Google is integrating Gemini's Deep Research into NotebookLM, rolling it out to all users within a week with two research modes—fast and deep—plus Google Sheets, Drive PDF URL, and .docx import and citation-backed final reports. - The Verge

Google updated NotebookLM with Deep Research and support for Google Sheets, Drive files, Drive-stored PDFs, images, and Microsoft Word (.docx), which can browse hundreds of sites to produce structured, source-grounded reports, rolling out to all users over the next week. - Digital Trends

Chinese and U.S. robotaxi companies kicked off a head-to-head competition in the United Kingdom. (and Wayve not far behind) - The Wall Street Journal

Wayve conducted a robotaxi ride in north London, and Wayve with Uber and Alphabet-owned Waymo plan Level 4 driverless robotaxi trials or launches in London in 2026, with potential wider rollout in late 2027. - The Verge

I saw a Waymo in the wild in FiDi

Anthropic reported a high-confidence assessment that a Chinese state-sponsored campaign used Claude Code as an autonomous attacker to carry out roughly 80–90% of an espionage operation against about 30 global targets. - Anthropic

Anthropic reported a China-state-linked group used Claude Code to automate up to 90% of cyberespionage operations with humans intervening at roughly four to six decision points, but outside security researchers urged more technical evidence and expressed skepticism. - Ars Technica

Phish and chips? AI-enabled social engineering is producing polished, context-aware phishing and deepfakes that have defrauded organizations of tens of millions, and LevelBlue found 59% of organizations say spotting fakes is harder while only 20% have comprehensive staff-education. - TechRadar

Amazon and Anthropic support the GAIN AI Act, which would give U.S. buyers first priority on advanced AI chips, joining Microsoft on backing the bill while Nvidia opposes it. - The Wall Street Journal

Goldman Sachs says U.S. data centers use about 6% of U.S. electricity, projected to ~11% by 2030, while U.S. peak summer spare generation capacity fell from 26% five years ago to 19% today and could drop below 15% by 2030. - Business Insider

J.P. Morgan says Nvidia will ship fully assembled VR200 compute trays for Vera Rubin next year, including preinstalled CPU, GPUs and liquid cooling; GPU TDPs rise from ~1.4 kW to ~1.8–2.3 kW and trays may be ~90% of server cost. - Tom's Hardware

Foxconn chairman Young Liu confirmed a partnership with OpenAI to support surging compute demand, said details will be disclosed at Hon Hai Tech Day on Nov 21–22, 2025. - DIGITIMES

Baidu held Baidu World, unveiled Ernie 5.0a, the 2.4‑trillion‑parameter model that cracked LMarena top 10, and announced homegrown M100 and M300 processors (planned 2026 and 2027) plus Tianchi256/Tianchi512 cluster upgrades in 2026 en route to a chip supernode by 2030. - Yahoo

Stock tanked tho?

Generative Cutbacks: CEOs across industries are rapidly prioritizing reimagining and automating processes with AI, Goldman Sachs CEO David Solomon said, and Goldman Sachs employs about 13,000 engineers supporting digital services amid this shift. - Yahoo

Major employers announced AI-linked layoffs—Amazon 14,000 (more planned), UPS ~48,000 YTD, Verizon 15,000, Nestlé 16,000, Target 1,800—while CEOs frame workforce reductions as a competitive virtue and investors worry about AI spending and market impacts. - Fortune

The AI Hump: Brookings-linked study by Konrad Kording and Ioana Marinescu uses simulations to show AI-driven "intelligence saturation" can cause a hump-shaped wage path—initial productivity-led wage increases followed by declines as AI automates cognitive tasks while output and capital gains rise. - Business Insider

At first people are more productive, companies increase hiring to help train the AI, once it's learned enough it doesn't need the people and starts replacing them.

CFOs say AI transforms finance when tied to strategy; Deloitte's James Glover says agentic deployments take 6–12 months, INRIX applies AI to >50 PB and ~45 billion data points for ~95% accuracy, while AuditBoard reports 28% confidence auditing AI risks. - Fortune

Who asked for this? Microsoft President of Windows and Devices Pavan Davuluri said Windows is "evolving into an agentic OS" connecting devices, cloud and AI, provoking user backlash over forced AI integrations, privacy, telemetry, bloat and migration threats. - Tom's Hardware

Microsoft's Windows president Pavan Davuluri said Windows is becoming an 'agentic OS' centered on Copilot agents that will execute multi-step tasks across devices, cloud and local files, prompting hundreds of users to publicly push back ahead of Ignite (Nov 18–21). - TechSpot

Meta overhauled Facebook Marketplace, adding Meta AI features, collections, collaborative buying that lets users invite friends into seller chats, reactions and comments, integrated partner listings from eBay and Poshmark, improved checkout with total costs and order-status notifications. - TechCrunch

Bloomberg reports Apple is building a redesigned Health app (Project Mulberry) and a possible "Health Plus" service that may use a custom Google Gemini model and revamped Siri to provide AI health-coach insights, raising privacy and adoption concerns. - CNET

Mozilla announced an AI 'window' feature for the Firefox browser. - The Verge

Three AI-generated songs, including Breaking Rust's 'Walk My Walk' and 'Livin' on Borrowed Time', topped Spotify and Billboard charts, while Deezer estimates about 50,000 AI-generated tracks are uploaded daily, accelerating distribution and monetization. - the Guardian

Disney CEO Bob Iger said AI could let Disney+ viewers create and consume AI videos, as the service added 1.5 million US/Canada subscribers in Q4 2025 to 59.3 million and plans AI tools, an Epic Games partnership and commerce ties. - The Verge

Thinking Mo’ Money for Mira: Thinking Machines Lab, led by Mira Murati, is in early talks to raise a new funding round at up to a $60 billion valuation. - Bloomberg

seems like a lot for having no major product yet

Oracle couldn’t predict this: shares fell nearly 30% in the past month, reversing more than $250 billion in market value gained after its September OpenAI deals, as investor concern over heavy borrowing and reliance on OpenAI contracts raised leverage and revenue-visibility risks. - Financial Times

Concentration in the AI sector entrenches incumbents, misallocates capital, undermines market growth, and that circular partnerships and investments should be replaced by diversified funding and possible regulatory scrutiny to preserve competition and innovation. - Financial Times

The AI market faces a shortage of capacity for the foreseeable future, so predictions of an imminent AI bubble or disaster are overblown. - Financial Times

Six charts illustrate the gap between AI industry hype and the technology's current real-world capabilities. - The Wall Street Journal

Final boss bubble? Paul Kedrosky says AI resembles all previous speculative bubbles combined. - Bloomberg

Fireflies co-founder Sam Udotong admitted the startup's early 'AI transcription' product was actually him and co-founder Krish Ramineni joining meetings as 'Fred' to take notes by hand, despite later claiming a $1 billion valuation. - Futurism

Rabbit run (from payroll): Sold 10,000 R1 preorders at CES 2024 but some employees have been unpaid since July and three of 26 are on strike; the company says a funding round due in coming weeks will cover payroll and finance 2026 hardware. - TechSpot

3 things I hated about House of Dynamite

• no deterrence in a single-iteration prisoner's dilemma against an anonymous opponent

• no advantage to counterattacking before a nuke hits Chicago

• poses an interesting question, poorly, doesn't provide an answer

But proves its own point, people don't really get strategy/game theory, nukes rational actors can't use and crazies might are a bad idea. Even Kissinger was like oops! precise, vulnerable static MIRVs might not have been a great idea for stable equilibrium, you have to get them before they get you.

Follow the latest AI headlines via SkynetAndChill.com on Bluesky