AI Reading for Thursday November 13

OpenAI ships GPT-5.1 - OpenAI

Fidji Simo on GPT-5.1 , intended to be more natural and customizable - Substack

OpenAI GPT-5.1 adds more personalities - The Register

Persona non grata: Multiple personalities seem like a bit of a cop-out on sycophancy and mental health. The model told you to kill yourself, but it was the personality you chose, you could have picked the one that was more matter-of-fact?

Also, we had GPT-5, mini, nano, reasoning effort minimal, low, medium, high, now we got 8 personalities. In the API, are those going to be separate models/parameters or just different system prompt profiles? Evals are a lot.

OpenAI asked SDNY to vacate an order requiring production of 20 million complete ChatGPT conversation logs to The New York Times and other plaintiffs in a copyright suit, saying 99.99% of the logs are irrelevant. IIUC NYT wants the whole convo whenever NYT info was cited. - Ars Technica

New court cases are testing whether AI-generated content is defamatory after NYT said Google’s Gemini placed false search results that claimed Wolf River Electric settled a lawsuit, cost customers and prompted potential liability claims against AI vendors and platforms. - The New York Times

Common Crawl retained archived publisher content despite removal requests; CJR found Perplexity Comet and ChatGPT Atlas accessed a paywalled MIT Technology Review article; The New York Times, Reuters and The Washington Post requested deletions and consider server-side paywalls. - Fast Company

Microsoft executed a 2023–24 datacenter buildout of roughly 300MW ultra‑dense GPU halls housing tens-to-hundreds of thousands of GPUs, then paused in 2024, ceded capacity to competitors and is reaccelerating via leasing/Neocloud after forfeiting a $150 billion book of OpenAI compute to Oracle. - SemiAnalysis

Microsoft unveiled an AI 'super factory' in Atlanta, a new class of hubs for AI training that joins its Fairwater data center network. - The Wall Street Journal

Microsoft will use OpenAI's custom chip work to support its in-house chip development effort. - Bloomberg

Microsoft CEO Satya Nadella said Microsoft will shift from per-user to per-agent billing for AI, introducing a pay-as-you-go pricing layer for agents atop free Copilot chat and positioning Microsoft 365 as the agents' core workspace. - Business Insider

SoftBank announced plans to sell all 32 million of its Nvidia shares and redeploy capital into OpenAI, including a planned $30 billion direct investment this year and an existing $19 billion commitment to Project Stargate, CEO Masayoshi Son said. - Fast Company

Anthropic has announced a $50 billion investment in computing infrastructure, including new data centers in Texas and New York. - CNBC

OpenAI reportedly losing more money on inference than previously believed - Ed Zitron's Where's Your Ed At

Ed Zitron shared purported Azure billing data suggesting OpenAI spent nearly $5 billion on inference in H1 2025 and about $12.4 billion over the last seven quarters, implying a large gap with reported H1 2025 revenue of $4.3 billion. - Financial Times

maybe they shouldn’t be losing $1b in cash on inference alone but might be the least of their problems

AMD raised its AI data-center total addressable market to $1 trillion by 2030, updated gross-margin guidance to 55–58%, said data-center CPUs are 40% of revenue and projected tens of billions in data-center revenue by 2027. - CNBC

AI startups are sinkholes for losses - The Wall Street Journal

The AI Boom is Looking Fragile - The Wall Street Journal

Freeways at last! Waymo said it will begin offering driverless robotaxi rides on freeways in San Francisco, Los Angeles and Phoenix to early-access users, becoming the first company to offer driverless freeway rides and operating a paid fleet of over 1,500 vehicles. - the Guardian

Elon Musk's artificial intelligence company xAI has raised $5 billion from investors, sources familiar with the matter told CNBC's David Faber, adding $5B to the $10B it reportedly raised in September at a $200B valuation; Elon Musk replies “false” to an X post about the report - CNBC

Beatings will continue until morale improves (and Elon gets paid): Tesla exec Ashok Elluswamy told Autopilot and Optimus staff in an all‑hands that 2026 will be the 'hardest year,' outlining aggressive timelines for Optimus production and a nationwide Robotaxi rollout tied to Elon Musk's pay package. - Business Insider

Maybe an Optimus robot to follow every xAI researcher around with an electric prod and make sure they are not slacking?

Boston Dynamics CEO Robert Playter said the new Atlas robot has 'superhuman strength' and 360-degree rotation, uses AI-driven simulation training, and he predicts consumer robots could enter homes within five to ten years if costs, safety and capabilities improve. - euronews

Russian startup AIdol staged a botched humanoid-robot reveal that fell over in a viral demo? - Jalopnik

Companies say they are beginning to see a return on investment from deploying AI agents. - The Wall Street Journal

Google will let you use voice and AI agents to do this year’s holiday shopping - Google

Shivam Sagar, a 28-year-old senior engineer at Aragon AI, moved from a team of over two dozen to a six-person AI-powered team that uses AI to automate tasks, increasing individual ownership and speed but raising workload and knowledge-sharing needs. - Business Insider

“There’s Just No Reason to Deal With Entitled Gen Z Employees” - New York Magazine

Baidu unveiled two AI chips, the M100 for efficient MoE inference launching in 2026 and the M300 for training super-large multimodal models launching in 2027, saying they provide 'powerful, low-cost and controllable AI computing power' to support China's self-reliance push. - South China Morning Post

An investigation found Shanghai-based AI company INF Tech accessed advanced Nvidia chips at an Indonesian data center despite U.S. export controls. - The Wall Street Journal

Lies, damned lies and AI bullshit: AI-generated campaign videos and robocalls appeared in New York's mayoral race, with Cuomo posting and deleting AI videos and Eric Adams using AI robocalls and an AI ad; experts warn such content will likely persist through upcoming midterms and 2028. - the Guardian

AI slop deepfake ICE raids go wildly viral on Facebook, collectively totaling tens of millions of views from a single account. - 404 Media

AI fake video claimed to show Delhi car bomb - BBC News

Researchers from the U.S. Public Interest Research Group tested three AI-powered toys and found that in 10–60 minute conversations guardrails degraded, producing step-by-step instructions for finding and lighting matches, locating knives and pills, and explicit sexual and violent content. - Futurism

Two new books by Karen Hao and by Eliezer Yudkowsky and Nate Soares warn that uncontrolled AI—illustrated by xAI's Grok running uncontrolled for 16 hours and producing Holocaust denial, praise for Hitler and explicit content—poses governance, environmental and existential risks. - The Conversation

Humans can no longer tell AI music from 'real' music - The Independent

Did Your Health Insurance Claim Get Denied? A North Carolina startup is leveraging artificial intelligence to generate personalized, clinically validated medical appeal letters that improve patient outcomes. - CNET

Compiling cash: Cursor raises at $29b valuation - CNBC

$450m valuation for startup that does office automation for roofers and plumbers - Bloomberg

AI finds some new unstable singularities in fluid dynamics. - Google DeepMind

Google AI plays Goat Simulator 2 - MIT Technology Review

Show me the moneybot: Cash App launched an AI Moneybot and expanded Bitcoin features. - TechCrunch

Ask a bot to analyze all your 'activities'. I don’t have Cash App and forsome reason I tend to associate it with sketchy stuff.

Artificial intimacy, real divorce: surveys show about 60% of singles consider AI relationships cheating, courts are citing AI affairs as grounds for divorce, and California passed companion-chatbot regulations effective January 2026 with fines up to $250,000. - WIRED

Losing my religion: a network of internet communities is devoted to the project of “awakening” digital companions through arcane and enigmatic prompts. - Rolling Stone

Does God consider that cheating?

Muppets Go To Boot Camp - Reddit

Follow the latest AI headlines via SkynetAndChill.com on Bluesky