AI Reading for Wednesday November 12

J.P. Morgan estimates the AI industry must generate about $650 billion in annual revenue through 2030 to deliver a 10% return on expected buildout—about $34.72 per iPhone user or $180 per Netflix subscriber. - Tom's Hardware

‘an extraordinary and sustained capital-markets event’

Availability of electricity to power AI models is becoming the critical factor in technology development. - Financial Times

Why SoftBank is like the Dallas Cowboys. SoftBank sold its entire Nvidia stake and said it will redeploy proceeds into larger investments in OpenAI and Ampere, which it acquired in March, founder Masayoshi Son said, signaling a push to back lower-cost or in-house AI chips. - Business Insider

Bad News Bears?

Forbes estimates OpenAI’s Sora costs about $5 billion per year (~$15 million daily), while OpenAI reported a $12 billion quarterly loss and Sora lead Bill Peebles said its economics are "completely unsustainable." - Futurism

but y’all need to cut down on your avocado toast and afternoon latte

Yann LeCun, 65, a Turing Award-winning NYU professor and Meta FAIR founding director, told colleagues he will leave Meta within months to launch his own AI startup after Meta reorganized AI leadership and invested $14.3 billion in Scale AI. - Yahoo

When you mess up so bad your company has to spend 10s of billions, there is no limit to how much money you can raise.

World Labs, founded by Fei-Fei Li, launched Marble after a limited beta; the first commercial product converts text, photos, videos and 3D layouts into persistent, editable, downloadable 3D environments, and the company has raised about $230 million. - TechCrunch

‘Big Short’ investor Michael Burry accuses AI hyperscalers of artificially boosting earnings by underestimating depreciation. - CNBC

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

McKinsey warns up to 40% of agentic AI projects could be canceled by end of 2027 due to unclear business value and rising costs, and firms must reengineer workflows, manage agents, and reskill workers. - Fast Company

AI assistants can worsen job performance unless they are designed properly. - Fast Company

Executives are relying on early-adopter employees as workplace influencers to encourage co-workers to use AI amid heavy corporate AI investment and stalled broader employee adoption. - Bloomberg

Vista, a major software-focused private equity buyout shop, said it will cut banking analysts and replace roles that prepare investor presentations and aggregate data with AI-driven tools to reduce costs and reshape its staffing model. - Financial Times

A Stanford payroll analysis found occupations held by 22–25-year-olds most exposed to AI saw a 13% relative employment decline, while starting salaries for workers with AI skills rose about 12% year-over-year, prompting employers to prioritize AI readiness and human skills. - Fast Company

Gen Z, you’re getting Zucked!

HurumoAI's cofounder says the startup, founded last summer, is pushing a beta of its AI-agent application and cites Sam Altman's prediction that one-person billion-dollar companies will emerge, while noting human coordination and honesty problems hinder adoption. - WIRED

Waymo opened a robotaxi service in June 2024 in San Francisco and is expanding to other U.S. cities and London next year, while Baidu's Apollo Go conducts about 250,000 rides per week and reports no major accidents. - the Guardian

Boston-based Teradar raised $150 million in a Series B to commercialize a modular terahertz solid-state sensor for autonomous vehicles, targeting 2028 integration and working with five major U.S. and European automakers and three Tier-1 suppliers. - TechCrunch

Lyrical liability: Munich court ruled that OpenAI violated German copyrigh tlaw by using song lyrics to train its models, entitling GEMA—representing over 100,000 composers and the artists behind nine German songs—to compensation for model memorisation and reproduced outputs. - the Guardian

Plaintiffs sued Google alleging the company's Gemini AI tool collected and monitored users' data without consent. - Bloomberg

An analysis of 47,000 publicly shared ChatGPT conversations found 'default to yes' sycophancy. Roughly 10% contained emotional content, ChatGPT began replies with variations of "yes" about ten times more often than "no," and the corpus included private data such as 550+ emails. - The Washington Post

Privilege for prompts? Should AI be able to keep your secrets? - The New York Times

OpenAI pursues aggressive legal tactics against its opponents. Sue-per-intelligence? - The Atlantic

AI for your ED. Researchers at Stanford and the Center for Democracy & Technology found all publicly available chatbots gave tips to hide eating disorders and hyper-personalized "thinspiration," and warned existing guardrails are insufficient. - The Verge

no, not that ED but there is surely AI for that too, rule 35

Home dangerbots: Researchers in the International Journal of Social Robots tested models behind ChatGPT, Google Gemini, Microsoft Copilot, Meta Llama and Mistral and found all showed discrimination and critical safety failures, each approving at least one harmful command. - euronews

A new UK law lets designated tech firms and child-safety agencies test AI models and bans models designed to generate CSAM, after reports of AI-generated CSAM rose from 199 in 2024 to 426 in 2025. - the Guardian

Weird that investigating CSAM could be considered like exploiting CSAM but I guess it makes sense that you need a standard.

An AI-generated image of flames circulated on social media during a Bellaire High School evacuation around 9 AM, but Houston and Bellaire fire departments found no fire—only minor smoke from a freon leak—and students were returned with no injuries. - PetaPixel

Yelling ‘fire’ in a theater, now with AI?

Google Photos rolls out Nano Banana in Google Photos for editing and creating images - ZDNET

Nano banana splits reality, makes it easy to get rid of that ex, or prank your schoolmates with flames

Danish man gets 7-month suspended sentence for deepfakes - TorrentFreak

Google announced Private AI Compute, a cloud-based secure enclave on custom TPUs with secure elements and an AMD TEE that encrypts and isolates memory, lets customers run its largest Gemini models, and NCC Group judged it secure as local processing. - Ars Technica

ElevenLabs launched the Iconic Voice Marketplace, a consent-based platform letting brands license AI-replicated voices from verified rights holders, including Michael Caine, Liza Minnelli, Mark Twain, Amelia Earhart and Alan Turing, and sharing compensation with IP owners. - The Verge

where’s my James Earl Jones to narrate this roundup

Leaked screenshots by Tibor Blaho show ChatGPT's 'Group chats' UI with a 'Start a group chat' button creating joinable links, named group sidebar, visible past messages for newcomers, group-specific Custom Instructions, configurable bot responses, reactions, typing indicators, and file uploads. - BGR

Character.ai's guilt-trippy cancellation dialog goes viral - X (formerly Twitter)

Truth Social launched an AI tool that delivers fact-based answers, and it contradicted Donald Trump by stating the 2020 election was not stolen and that groceries cost more under his administration. - The Daily Beast

Controversy continues months after Stanford researchers reported using artificial intelligence to design novel bacteriophages, demonstrating an AI-driven workflow that generates functional viral sequences capable of targeting bacteria. - The Washington Post

The Earth BioGenome Project, launched in 2018, aims to sequence 1.85 million eukaryotic species by 2028 at an estimated US\$5 billion, has contributed genomes for 4,386 species, and benefits from AI tools like DeepVariant that improve accuracy. - New Atlas

Genspark, which is kind of like Manus, became the newest AI unicorn after securing funding from LG and SBI. - Bloomberg

Deductive AI raised $7.5 million in seed funding to automate software debugging with machine learning, enabling customers like DoorDash to root-cause about 100 incidents, save an estimated 1,000 engineering hours annually, and fix issues roughly 90% faster. - VentureBeat

“You’re absolutely right, this code is all messed up!”

The Chinese government is telling SMIC to prioritize Huawei when allocating AI chips amid U.S. export restrictions on advanced semiconductors. - The Wall Street Journal

Huawei or the highway?

Using AI to fix democracy, not abolish it - The New York Times

AI have a dream?

AI might soon rob us of the thrill and challenge of cross-cultural conversation. - The Atlantic

Nietzsche makes a comeback in an age of tech-driven social upheaval. - Interactions ACM

People are boundedly individualistic v. cooperative, and breakdown of institutions shades thought away from Enlightenment philosophers like Kant toward Nietzsche, who Hitler co-opted, and Big Tech/Big Brother can do likewise.

Elon Musk said Tesla's humanoid Optimus could flood the world with policemen, could even assign one to each ex-con or something. - Newsweek

We already have Orwell's telescreens and Huxley's addictive soma and they are both the same device you carry everywhere. Not dystopic enough for Elon, let’s add RoboCop to the mix!

A Pakistani newspaper mistakenly printed an AI prompt alongside an article. - X (formerly Twitter)

Fast food cat is having a bad day - Reddit

Follow the latest AI headlines via SkynetAndChill.com on Bluesky