AI Reading for Thursday October 30

OpenAI plans an IPO that could value it up to $1 trillion, has eyed raising about $60 billion, may file in H2 2026 for a 2027 listing, and the OpenAI Foundation and Microsoft hold about 26% and 27%, respectively. - Reuters

SEC filings show Microsoft funded $11.6 billion of a $13 billion commitment, holds a 27% stake and recorded a $3.1 billion reduction in net income this fiscal year, implying an approximately $11.5 billion quarterly net loss at OpenAI. - The Register

OpenAI is paying more than 100 former investment bankers about $150 per hour (≈$300,000/year) to train GPT models to perform junior-analyst tasks such as discounted cash-flow analysis, deal rationale and pitchbook assembly. - Quartz

OpenAI updated Sora 2 to add reusable "character cameos" avatars (people, pets, illustrations), clip-stitching for multi-scene videos, and leaderboards, with per-character permissions and limited testing in the US, Canada, Japan and Korea amid a Cameo trademark lawsuit. - The Verge

SDNY judge preserves authors' download claim against OpenAI, suit now limited to GPT‑3 through GPT‑4o Mini - Decrypt

U.S. District Judge Sidney Stein allowed two additional infringement theories in George R.R. Martin and other authors' suit against OpenAI to proceed, affirmed that ChatGPT outputs can be substantially similar, and said statutory damages reach up to $150,000 per work. - The Hollywood Reporter

An AGI narrative popularized by Ben Goertzel and Shane Legg, and amplified by LessWrong, Eliezer Yudkowsky and Nick Bostrom, is reshaping AI investment and strategy, driving OpenAI’s roughly $500 billion valuation and multibillion infrastructure deals with Nvidia and AMD. - MIT Technology Review

Researchers used three open-source AI packages to generate about 75,000 protein variants from 72 toxins (starting with ricin), and four commercial DNA-order screening programs showed wide detection variation, prompting three vendors to issue updates that improved detection. - Schneier on Security

Paul Scharre says autonomous weapons—already including automatic air-defence interceptors, Israel's Harpy loitering munition and Ukrainian ML-guided terminal systems—are risking an arms race as autonomy enables swarming, persistent loitering and autonomous strikes. - Nature

Alphabet reported record Q3 revenue of $102.3 billion, up 16% year-over-year, with Google Services $87.1 billion, Google Cloud $15.2 billion (up 34%), AI Mode 75 million daily active users and Gemini 650 million monthly active users. - Adweek

AI-driven updates lifted Q3 revenues for Alphabet to $102.34 billion and Meta to $51.25 billion, prompting both to sharply raise 2025 capex guidance—Alphabet to $91–93 billion and Meta to a $70 billion floor and boost server spending. Google's stock rose, Meta's fell sharply. - The Register

Google agreed to a secret 'wink' disclosure canary if governments sought Israeli cloud data - the Guardian

YouTube CEO Neal Mohan announced an AI-focused reorganization offering voluntary buyouts and splitting the product organization into viewer, creator and community, and subscription groups to accelerate AI-driven product development and monetization. - Engadget

YouTube does AI rescaling but lets you opt out. - The Verge

DeepMind's AI hurricane model, trained on nearly 5,000 storms over 45 years, has been used operationally by the NHC and predicted Hurricane Melissa's rapid intensification, raising Cat-5 odds from about 50–60% on Oct 21 to ≥80% on Oct 23. - Nature

DeepMind-led 75-page study "Generative Chess Puzzles" trained large neural networks on 4 million Lichess puzzles and used reinforcement learning to generate creative chess problems evaluated by GMs Matthew Sadler, Jonathan Levitt and FM Amatzia Avni. - Chess.com

Google DeepMind and Google.org launched the AI for Math Initiative with five partner institutes, providing funding and access to DeepMind tools (Gemini Deep Think, AlphaEvolve, AlphaProof) to accelerate math discovery, including a 48-multiplication 4x4 algorithm and IMO-level results. - Google

Investors seem to tire of Meta's big spending despite higher than expected earnings. Q3 capex doubled to $19.4 billion with full-year guidance raised to as much as $72 billion and Reality Labs lost $4.4 billion. - Quartz

Meta plans to spend $70–$72 billion on AI infrastructure this year and expects capex to rise through 2026 to build data centers, cloud services and hire researchers for its Superintelligence Labs, CFO Susan Li said; stock dropped - Business Insider

Meta says its Hyperion AI data center in Richland Parish, Louisiana projects $27 billion, up from $10 billion announced in late 2024, and will be financed through an 80/20 joint venture with Blue Owl Capital as Meta leases the campus. - NOLA.com

Meta moved to dismiss Strike 3's suit alleging it torrented about 2,400 adult films via corporate and roughly 2,500 hidden IPs from 2018–2024, denies using the material to train AI and says Strike 3 seeks over $350 million. 'Personal use' by whom? - Ars Technica

Recent Fed rate cuts and markets pricing in more easing could further inflate already stretched AI stock valuations. Analysts warn of a bubble, with the Magnificent Seven comprising about 35% of the S&P 500 and the index P/E near the dot‑com peak. - Investopedia

AI is the final boss bubble - WIRED

AWS opened an $11 billion Project Rainier data center campus in Indiana for Anthropic, with seven buildings housing nearly 500,000 Trainium2 chips (expected to exceed one million by year-end) and plans to add 23 buildings to exceed 2.2 GW capacity. - SiliconANGLE

If you artificially manipulate the LLM's activations and ask it if it notices anything, it kind of does? - Anthropic

A.I. Threatens Our Ability to Understand the World - The New York Times

Larissa Schiavo of Eleos AI is "preparing for AI sentience and welfare." - Bloomberg

If you write 'wash me' on a dirty truck, it's probably not sentient?

1X’s Neo wants to be your housekeeper. First, it needs to be controlled by a human in your home. Cool with you? - The Wall Street Journal

AGI = Actually A Guy In India who now gets to be a voyeur - Reddit

Universal Music Group is partnering with AI music generator Udio to launch a 2026 subscription service that lets users create music with licensed songs, partly as a means to settle a copyright lawsuit. - The Wall Street Journal

Apparent AI Hallucinations in Filing from Two >500-Lawyer Firms, who are ordered to propose sanctions reflecting "the seriousness of its use of fake citations and appropriate steps to ensure that this type of violation will not occur again" - Reason.com

AI-generated deepfake videos impersonating physicists Michio Kaku and Brian Cox are spreading on social platforms, falsely claiming third-detected interstellar object 3I/ATLAS is an alien spacecraft; Kaku and scientists are urging YouTube, TikTok and AI firms to remove them. - PCMAG

A deepfake Jensen Huang outperformed Nvidia's real GTC keynote on YouTube – with five times the viewers – promoting a crypto scam, natch - TechSpot

Are large language models making us less productive? - Financial Times

EA employees say management-mandated internal chatbot 'ReefGPT' hallucinates and generates buggy output, increasingly their workload as they fix its output. - PC Gamer

But AI said… - Stop Citing AI

Grammarly Inc. rebranded as Superhuman, consolidating Grammarly, Coda and Superhuman Mail into the AI-native Superhuman Suite and launching Superhuman Go—an AI assistant connecting 100+ apps—plus an Agent Store and closed-beta Agents SDK, available today on paid plans. - SiliconANGLE

GitHub wants to be the Eisenhower of agents, supreme commander of your army of Copilot, Claude, Cursor and other agents. - The GitHub Blog

Character.ai to ban under-18s after first limiting hours to gradually wean them off, lest they go into withdrawal and do something foolish. - Financial Times

Testing five AI browsers—ChatGPT Atlas, Perplexity Comet, Kosmik, Dia and BrowserOS—each offering different features, platform support and subscription or open-source pricing. - ZDNET

Insilico's AI‑discovered longevity drug enters trials - Fortune

Nabla, launched in 2023 and used by hundreds of US healthcare organizations, uses ambient AI to record and summarize encounters with high-90s accuracy, reducing documentation that the AMA says consumes at least eight weekly hours and contributes to physician burnout. - MIT Technology Review

Using Claude AI, a grieving family who paid $20/month obtained an itemized hospital bill and reduced a $195,000 charge to $33,000 by identifying about $100,000 in duplicative charges, improper coding and disputed ventilator billing. - Tom's Hardware

Recruiters are ‘drinking through a fire hose’ of job applications, and blame AI - CNBC

AI is everywhere, as the author learned the hard way while making as little contact as possible with machine learning and generative artificial intelligence. Software is eating the world, and AI is eating software. Unless you go to a desert island, forget it. - The New York Times

KitKat, a bodega cat in SF's Mission district, was found dying Monday night; owner Mike Zeidan and neighbors say a Waymo autonomous vehicle hit the cat, and Waymo has not offered independent confirmation. - SFist - San Francisco News, Restaurants, Events, & Sports

Man pranks ChatGPT by speaking to it in hex and 'makes it sweat' - Supercar Blondie

Follow the latest AI headlines via SkynetAndChill.com on Bluesky