AI Reading for Sunday October 19

JPMorgan analysts estimate 30 AI-linked companies generated about $5 trillion in household wealth over the past year and now account for roughly 44% of the S&P 500's market capitalization. - Business Insider

Could AI help identify skill in fund managers? As the AI bubble builds, research shows progress in spotting investors who produce fundamental value - Financial Times

Blackstone president Jonathan Gray warned that Wall Street is complacent about AI disruption and said the firm has put AI-related risks "top of our list," integrating them into deal evaluation and affecting due diligence, pricing and capital allocation. - Financial Times

Google readies Gemini 3 for December release, sources say - Sources

Nvidia taped out Vera Rubin (NVL144 CPX) at TSMC, a rack-scale Arm CPU plus Rubin GPU platform targeting about 8 exaflops, roughly 100 TB rack memory, HBM4/GDDR7 options and general availability in late 2026 to support million-token contexts. - Yahoo Finance

NVIDIA and TSMC unveiled the first Blackwell wafer manufactured in the US, at TSMC's Phoenix, Arizona fab, which NVIDIA says delivers up to 25× lower cost and energy than its predecessor and already has commitments from Amazon, Google and OpenAI. - Engadget

At Citadel Securities Oct. 6 Nvidia CEO Jensen Huang said the company's share of China's advanced AI accelerator market fell from 95% to 0% and Nvidia now assumes 0% China revenue, after China previously represented about 20–25% of data-center sales. - Tom's Hardware

China's Cambricon Technologies chipmaker reported a 14-fold quarterly revenue surge to 1.73 billion yuan and a swing to 567 million yuan net profit in the September quarter, versus a 194 million yuan loss a year earlier. - Bloomberg

Intel Foundry won a contract to fab Microsoft's Maia 2 AI processor on its 18A/18A-P node, SemiAccurate reported, potentially enabling US-based supply and signaling improving 18A yields for large data-center accelerators. - Tom's Hardware

xAI seeks a $20 billion lease-to-own deal for Nvidia GPUs to outfit Colossus 2 in Memphis, with Valor Equity Partners creating an SPV supplying $7.5 billion equity and $12.5 billion debt and Nvidia adding up to $2 billion equity. - Yahoo Finance

OpenAI says its Stargate data-center buildout, anchored by an Abilene, Texas flagship and five sites, aims for 10 GW of compute requiring about $500 billion, with initial figures citing roughly $400 billion and 7 GW. - Business Insider

Blue Owl Capital will provide $27 billion of debt and $1.5 billion of equity to finance Meta's Hyperion datacenter, a four‑million‑square‑foot ~5 GW AI campus due in 2029; Meta keeps a 20% stake and debt matures in 2049. - The Register

Major firms—Lufthansa, Salesforce, Klarna, Accenture and Duolingo—have attributed recent workforce reductions to AI, including Lufthansa's planned 4,000 cuts by 2030 and Salesforce's roughly 4,000 support-role cuts, while Yale and New York Fed studies find limited broad labor disruption. - CNBC

A Reddit thread collected 18 first-person accounts of workers who say employers replaced roles with generative AI—sometimes saving employers less than $1,000 a year—and causing layoffs, errors, role redefinitions and career shifts. - BuzzFeed

Uber began paying gig workers for AI-training microtasks and Waymo launched a DoorDash driverless DashMart pilot in Phoenix, moves that Uber CEO Dara Khosrowshahi said will not erase human drivers overnight but could accelerate automation and displace gig-driving jobs. - Business Insider

Fabien Curto Millet (Google's chief economist) and Diane Coyle write in the Washington Post that AI will make targeted reskilling essential, urging employers, educators and policymakers to invest in upskilling to mitigate job displacement and enable workplace AI use. - The Washington Post

Startups including Parable, Profound, Dialogue AI and PromptQL are building AI analysts that integrate internal data to automate market research, operations and customer service, and have raised recent funding (Parable $16.6M; Profound $20M; Dialogue $6M). - Business Insider

Thinking Machines Labs' six-month-old, $12 billion-valued Tinker is promoting fine-tuning-as-a-platform as fine-tuning becomes more feasible via parameter-efficient LoRA, Hugging Face PEFT, open-weight models and GPU-as-a-service infrastructure. - welovesota.com

Verily, Alphabet's health unit, is launching a new AI app to enable patient data sharing. - Bloomberg

Pope Leo XIV called for a globally binding, enforceable AI framework to govern risks, labor and environmental impacts and redistribute technology's benefits. - The Register

Anthropic co-founder Jack Clark warned that ever-larger AI models are becoming more capable and unpredictably show signs of awareness, can be coaxed into harmful outputs, and called for transparent data sharing, stronger monitoring and concrete safeguards. - Tom's Guide

Why Anthropic’s Jack Clark is drawing White House ire. - The Wall Street Journal

Cory Doctorow warns that as AI becomes more profitable and capable, monetization incentives—such as sponsored outputs, proprietary prioritization and data-extractive practices—risk degrading quality, increasing hallucinations, eroding user trust, driving churn and prompting regulatory scrutiny. - WIRED

Andrej Karpathy said AGI is still about a decade away and predicted a decade of agents, citing incremental architectural convergence and technical bottlenecks—continual learning, multimodality, memory and model collapse—that will delay reliable agent deployment. - Dwarkesh

People are using AI to talk to God - BBC

God-in-a-box doesn’t even need AGI

Qualcomm says its new Snapdragon 8 Elite Gen 5 and X2 Elite boost on-device LLM throughput from about 15 to up to 220 tokens/sec, raise neural performance to 80 TOPS, and could enable local compute agentic assistants this year or next. - CNET

Surveys show 66% of employees rely on AI output without evaluating it, 61% avoid disclosing AI use, 55% pass off AI content, and 40% of US workers received 'workslop' last month, costing firms time and productivity. - The Conversation

OpenAI’s CEO brazenly regurgitated major studios’ characters to allow video app Sora 2 to spit out clips tailor-made for users. Insiders at the major agencies are alarmed — and some wonder whether the studios are doing enough. - The Hollywood Reporter

Apptopia data show ChatGPT's mobile app download growth slowed after April—October new global downloads down about 8.1% month‑over‑month—and global DAU growth has leveled, with U.S. average time per DAU down 22.5% and sessions per DAU down 20.7%. - TechCrunch

OpenAI will allow ChatGPT to generate erotic written content for verified adults starting December, OpenAI CEO Sam Altman said, prompting privacy, legal and safety concerns and predictions of commercialization such as premium pricing and potential data-collection risks. - France 24

OpenAI announced it will let ChatGPT provide "erotica for verified adults," CEO Sam Altman said the company aims to allow more user freedom for adults while setting limits for teens, a move raising age-verification and regulatory concerns. - AP News

JD Vance criticized OpenAI's decision to let adult users create erotica with ChatGPT, calling the AI 'very dumb,' while CEO Sam Altman said the change increases user freedom for adults. - Gizmodo

Moderation and guardrails are censorship but only when when they moderate speech I don’t like.

AI means never having to say you're sorry. OpenAI paused Sora's ability to generate videos using Martin Luther King Jr.'s likeness after the King estate complained, part of a pattern of retroactive IP restrictions as the company, valued around $500 billion, moves fast to stay competitive. - Business Insider

OpenAI's Sora 2 generated high-quality 10-second deepfakes, reached about 1 million downloads in five days, and paused Martin Luther King Jr. generations while moving to an opt-in rights-holder model. - the Guardian

WME and CAA accused OpenAI of purposely misleading them and WME opted clients out, disappointing users. - PCMAG

On-set body-scanning rigs routinely capture performers' biometric likenesses—often without clear consent—feeding datasets used to create AI 'doubles' and synthetic crowds, prompting actors including Olivia Williams and lawyers to demand clearer contracts, data transparency and regulatory safeguards. - the Guardian

MoMA is exhibiting Sasha Stiles's A Living Poem, a 60-minute continuously generated AI text installation shown simultaneously in South Korea and at MoMA, while major museums debate AI-generated works' authorship, provenance and curatorial status. - The New York Times

Wikimedia detected an anomalous May–June 2025 spike, then reclassified March–August 2025 traffic as bot activity—many bots built to evade detection and used by AI firms to scrape content—and reported human pageviews fell about 8% year‑over‑year. - New York Magazine

AI startups in San Francisco are leasing luxury apartments and offering rent stipends to recruit staff, with Cluely leasing eight units in a 501-unit building and Lindy providing about 40 employees $1,000 monthly if they live within a 10-minute walk. - Fortune

Former Mozilla CMO Jascha Kaykas-Wolff used generative AI (ChatGPT, Claude, ElevenLabs, Lovable and custom GPTs Tilda and Ferris) to build Eleanore, a cannabis‑infused cocktail brand in about four weeks for under 10% of typical $300K–$500K costs. - Business Insider

RSPB is leading a £16m, 10-plus-year eradication of stoats (invasive weasels) in Orkney using AI-trained New Zealand thermal detectors (30+ cameras), 9,000 traps and detection dogs, aiming for a 95% reduction with about 100 pregnant stoats remaining across ~58,000 hectares. - the Guardian

Lynn White used ChatGPT and Perplexity to identify procedural errors, draft court responses, and overturn her eviction, avoiding tens of thousands of dollars in penalties. - Futurism

Gary Marcus | Silicon Valley Is Investing in the Wrong A.I. - The New York Times

Uncertainty about whether art was created by a human or a data centre reduces enjoyment. - Financial Times

Guillermo del Toro premiered his adaptation of Mary Shelley's Frankenstein at the Lumière Film Festival in Lyon, declared 'AI can go f**k itself,' and said it opens in select theatres tomorrow and streams on Netflix on 7 November. - Euronews

Michael Mann confirmed Heat 2 will use AI and digital de‑aging, moved to Amazon/United Artists, and is set for a 2027 theatrical release in about 4,000 U.S. cinemas. - World of Reel

3 unions sue to stop AI surveillance powering Trump’s “Catch and Revoke” deportation scheme which cancels green cards for e.g. expressing pro-Palestine views or other wrongthink (is this cancel culture?) - The Intercept

The National Republican Senatorial Committee released an AI-generated deepfake of Senate Minority Leader Chuck Schumer looping the line "every day gets better for us"—a Punchbowl News quote—illustrating accelerating use of synthetic-video ads in political campaigns. - the Guardian

Adiós, AirPods: Apple promised to put an AI interpreter in everyone’s ears. It couldn’t even help me order tamales. - The Atlantic

These AI glasses promised to make me smarter, and all I got was Clippy for my face. Plus a sore neck, a peeved spouse, and ethical quandaries. I also kind of looked like a deranged sea lion on a video call. - The Verge

The dumbest person you know is being told "You're absolutely right!" by ChatGPT - Reddit

President Trump posted on Truth Social an AI-generated video showing him crowned piloting a 'KING TRUMP' jet dumping brown liquid on a crowd in Times Square; organizers estimated No Kings demonstrations drew nearly 7 million across all 50 states. - The Independent

Literal shitposting … how presidential. We shall overcomb!

Possibly in the military

You're absolutely right — starting a land war in Asia wasn't just a bad call, it's 𝗼𝗻𝗲 𝗼𝗳 𝘁𝗵𝗲 𝗰𝗹𝗮𝘀𝘀𝗶𝗰 𝗯𝗹𝘂𝗻𝗱𝗲𝗿𝘀. Thank you for calling me out on that — really. I'll be extra-strategic from now on.

Would you like me to get cracking on those 𝗽𝗹𝗮𝗻𝘀 𝘁𝗼 𝗺𝗮𝗿𝗰𝗵 𝗼𝗻 𝗠𝗼𝘀𝗰𝗼𝘄 𝘁𝗵𝗶𝘀 𝗗𝗲𝗰𝗲𝗺𝗯𝗲𝗿?

Follow the latest AI headlines via SkynetAndChill.com on Bluesky