AI Reading for Thursday October 16

Goldman Sachs projects widespread generative AI adoption could add about $20 trillion to the US economy, boost US labor productivity roughly 15% over a 10-year adoption period, and sees about $300 billion spent annually on AI infrastructure in 2025. - Business Insider

Ten AI groups, including OpenAI and Elon Musk's xAI, have gained almost $1 trillion in value over the past year as unprecedented funding drives start-up valuations, prompting warnings of a possible bubble and concern about market sustainability. - Financial Times

The AI investment bubble poses a larger global economic threat than U.S. tariffs, and financial markets and economic realities have constrained America's use of import duties. - Financial Times

Wall Street warns an AI-driven market bubble as JPMorgan CEO Jamie Dimon calls elevated asset prices "a category of concern", BofA's fund-manager survey names an "AI equity bubble" with cash at 3.8%. - Yahoo Finance

JPMorgan Chase CEO Jamie Dimon said AI "isn't a bubble," citing claims that AI-related firms drove about 80% of US stock gains and warning that some costly infrastructure projects may fail or lack the power they need. - Business Insider

JPMorgan CEO Jamie Dimon said AI is real but some asset prices are "in bubble territory," and that JPMorgan has invested billions since 2012 with over 2,000 AI staff and reported roughly $2 billion in savings or revenue. - Yahoo Finance

Major banks accelerated AI adoption in Q3 to boost efficiency: Citigroup reported ~180,000 employees using proprietary AI and over 1 million AI-driven code reviews this year (≈100,000 developer hours weekly), and JPMorgan estimated about $2 billion in AI upside. - CIO Dive

TSMC raised its 2025 revenue guidance to mid-30% growth from around 30%, maintained 2025 capital spending guidance at up to $42 billion, and reported a record Q3 net profit up 39.1% to T$452.3 billion, citing strong AI-driven demand. - Bloomberg

Surging AI-driven data center demand has created a $400 billion global gas-turbine shortage, with lead times stretching seven to eight years and manufacturers like Siemens Energy and Mitsubishi hesitant to rapidly expand capacity amid past market volatility. - Yahoo Finance

AI companies are investing heavily in large-scale infrastructure assuming algorithms will keep improving with scale, but those improvements may not materialize, risking overinvestment. - WIRED

AI data-center operators, facing electricity shortages, are constructing on-site power plants to meet growing energy demand. - The Wall Street Journal

Oracle can now vectorize customer records to feed AI models and agents, Larry Ellison said, and announced an Abilene campus planned for about 500,000 Nvidia GPUs and 1.2 GW power, citing a $455 billion (cough-OpenAI) AI backlog. - The Register

Cisco's 2025 AI Readiness Index finds only 13% of companies AI-ready, 86% expect productivity gains within three years, and 54% say infrastructure can't scale for AI workloads, warning of widespread 'AI infrastructure debt'. - The Register

AI Startup Poolside Teams With CoreWeave on Massive Data Center - Bloomberg

We doing $14b pre-product startups with 2GW data centers now.

Meta will spend more than $1.5 billion to build a gigawatt-scale data center in El Paso, Texas, targeting 1 GW capacity to power high-end AI chips and expected to come online in 2028, the company said. - Yahoo Finance

Struck me that Google announced $15b for 'gigawatt-scale' data centers in India, OpenAI announced $25b for 500MW in Argentina, Meta announced $1.5b for a gigawatt in El Paso. There are different levels of capital intensity if you have to build all the power and infra, but anyway.

OpenAI announced apps inside ChatGPT, enabling brands to embed interactive services within conversations and turning ChatGPT (800+ million users) into a unified entry point for inline flows like booking, property maps, slides and music. - Hacker Noon

y tho, who asked for this

OpenAI reported H1 2025 revenue of $4.3B and a $13.5B net loss, while ChatGPT has about 800 million users but only roughly 5% pay, generating about 70% of OpenAI's recurring revenue. - The Register

Upgrades: OpenAI updated Sora 2 to let free users generate videos up to 15 seconds (previously about 10 seconds), while Pro subscribers ($200/month) can create up to 25‑second videos and access a Storyboard tool, the company said. - TechRadar

OpenAI will allow sexually explicit "erotica" on ChatGPT for verified adults beginning in December, CEO Sam Altman said, drawing criticism from the National Center on Sexual Exploitation which warned of mental-health harms and urged reversing the change. - Variety

Sam Altman says OpenAI isn't 'moral police of the world' after erotica ChatGPT post blows up - CNBC

OpenAI issued wide-ranging subpoenas tied to its lawsuit with Elon Musk to at least seven nonprofits critical of its conversion—including donor lists, communications and governance documents—prompting those groups and legal experts to challenge the scope and threaten motions to quash. - NBC News

OpenAI employee Leopold Aschenbrenner warned AI could match or exceed human capacity by 2027 and create a "permanent underclass," while MIT (95% negligible returns), Carnegie Mellon (~70% task failure) and the Bank of England flagged sustainability concerns. - the Guardian

Leopold Aschenbrenner, 23, a former OpenAI employee who began at FTX's philanthropy arm, now runs an AI-focused hedge fund managing over $1.5 billion, drawing both praise as an AGI visionary and criticism for his limited experience. - Fortune

Ke Yang left Apple to join Meta after being promoted to lead its AI-driven Answers, Knowledge, and Information (AKI) group, which is building the planned March 2026 Siri overhaul. - Bloomberg

Meta hired Andrew Tulloch, co-founder of Thinking Machines Lab, after offering a compensation package reportedly worth up to $1.5 billion over at least six years, Thinking Machines Lab said he has decided to pursue a different path. - Yahoo Finance

Anthropic aims to nearly triple annualized revenue in 2026, sources say - Yahoo Finance

Anthropic takes on OpenAI with Claude Haiku 4.5, a free model for Claude.ai, with paid API rates of $1 per million input and $5 per million output, 200,000-token context (1M for developers), ASL-2 safety classification, and competitive coding benchmarks. - VentureBeat

Anthropic released Claude Haiku 4.5, the newest version of its smallest model, billed as offering similar performance to Sonnet 4 "at one-third the cost and more than twice the speed." - TechCrunch

Claude Haiku 4.5 was released and is available across Claude API, Claude Code, apps, Amazon Bedrock and Google Vertex AI, priced at $1/$5 per million input/output tokens, delivering ~90% of Sonnet 4.5 coding performance and up to 4–5× faster speed. - Anthropic

Google's Gemini 2.5 Pro has led benchmarks in coding, mathematics, logic and emotional reasoning, topped blind user-preference rankings against 26 models, and is being integrated across Android, Google Drive, NotebookLM and Google Home while expanding to wearables. - Tom's Guide

If Google brings Gemini and its Search app to macOS, it could integrate web, Drive and local file search with conversational Cmd+Space responses and agentic task execution, directly competing with Apple Spotlight and exposing Siri and Apple Intelligence gaps. - Android Authority

Google released Veo 3.1, an upgraded video model arriving in the Gemini app and Flow filmmaking tool that offers stronger prompt adherence, improved audio and realism, support for landscape and portrait 16:9 outputs, and a lower-cost "Fast" variant. - Ars Technica

iOS 26 Apple Intelligence adds 20+ AI features including Live Translation and voicemail summaries, limited to newer iPhones - 9to5Mac

Apple unveiled an updated $3,499 Vision Pro headset with a new main chip and longer battery life, positioning it as a high-end "face computer" amid tepid consumer demand and intensifying competition. - The Washington Post

Cloudflare CEO Matthew Prince updated a Content Signals Policy that altered robots.txt on millions of sites to block use of pages for Google's AI Overviews, affecting roughly 20% of the web and aiming to protect publishers' referral revenue. - Ars Technica

DeepMind and Yale used a 27-billion-parameter Gemma-based model (C2S-Scale 27B) to predict and experimentally validate that CK2 inhibitor silmitasertib plus low-dose interferon increased MHC-I antigen presentation about 50% in human neuroendocrine cells. - Decrypt

Researchers from Stanford, Northeastern and West Virginia University propose a "verbalized sampling" magic prompt that asks a model to generate five responses with probabilities, a training-free inference-time method that increased output diversity about 1.6–2.1x in tests. - Decrypt

LimeChat says its AI chatbots cut staffing needs about 80% for 10,000 monthly queries, handled ~70% of complaints and helped revenue rise to $1.5M in 2024, while analysts forecast job and revenue losses in India's BPM sector. - Yahoo

In 2025 major employers including Nestlé (16,000 jobs), Chevron (~9,000), UPS (20,000), BP (7,700), Intel (5,000+), Starbucks (~2,000) and numerous tech firms are cutting thousands of jobs, with AI and automation cited as drivers. (I'm sure this is fine) - Business Insider

Amazon plans a 2025 global round of layoffs across divisions including PXT and AWS, continuing cuts of more than 27,000 jobs since 2022, with analysts estimating 10,000–15,000 roles at risk as the company shifts toward generative AI. - www.ibtimes.co.uk

Australia's Commonwealth Bank reversed its decision to make 45 customer service roles redundant after staff who trained AI chatbots protested, with chairman Paul O'Malley saying "We made a mistake" amid the bank's $300 million AI investment and union, regulator scrutiny. - au.finance.yahoo.com

Candidates are using generative-AI apps during live video job interviews, illustrated by viral TikTok clips (one with 5.3 million views), while employers' use of AI screening has prompted HR to consider detection and stricter proctoring. - The Atlantic

Western manufacturing executives who visit China are coming back terrified, as robotics has catapulted Beijing into a dominant position in many industries - The Telegraph

Ken Griffin says generative AI helps productivity but does not help investors beat the markets. - Bloomberg

JEDEC finalized the HBM4 standard, targeting ~8 Gbps per pin (≈2 TB/s per stack) and up to 64 GB per stack; SK Hynix and Micron sampled higher‑speed parts in 2025, SK Hynix held ~62% market share in Q2 2025. - The Register

AI-driven tools are enabling N=1 medicine—collapsing drug-discovery timelines from years to weeks, reducing costs once over $2.6 billion, with the AI-in-drug-discovery market projected to reach $20.3 billion by 2030 and companies like Nome advancing personalized therapies. - Forbes

PainChek uses smartphone facial analysis to output 0–42 pain scores, distinguishing between physical pain and emotional distress in patients in residential facilities, with pilots showing a 25% drop in antipsychotic use and 42% fewer falls; other AI tools report >80–90% accuracy. - MIT Technology Review

Wysa (~5 million users), Youper (~3 million) and general chatbots are used as on‑demand therapists; a study found 12% of American teens sought AI emotional support and an APA poll reported ~30% of psychologists used AI. - The Week

A think-tanker told Congress AI can democratize expertise and provide personalized learning but poses risks—academic integrity/over-reliance, pedagogical misalignment, and psychological harms—and recommended research, regulatory sandboxes, benchmarks, and system reporting. - American Enterprise Institute - AEI

An Oxford University Press survey of 2,000 UK 13–18-year-olds found 80% use AI for schoolwork, over 90% said it aided at least one academic skill, 60% reported it harmed other abilities, and 51% wanted clearer guidance. - Business Insider

CCTV's Financial Investigation reported Chinese paper mills using generative AI chatbots to mass-produce forged academic papers, with workers completing over 30 articles weekly and one Wuhan agency handling more than 40,000 orders annually. - Yahoo

Analysis of 114 U.S. court matters (June 2023–Oct 7, 2025) from Damien Charlotin's AI Hallucination Cases found 56.1% of alleged AI-tainted filings by plaintiffs, 30.7% by defendants, ~90% from solo/small firms, and ChatGPT named in 18 of 34 tool-identified cases. - Stanford CIS

Vendors are deploying "weapons-grade" AI to untrained users, contributing to high failure rates—citing "95 percent of junked AI proofs of concept"—and failures will continue until organizations adopt skills, governance and process changes. - The Register

What are the 2 categories of AI use and why do they matter? The divide   between builders and consumers  is shaping not only how generative AI is used but also whether it delivers value at all. Even if you are a consumer you need to be in the weeds and hands-on. - Fast Company

After seven years implementing AI, the author finds AI is a feature, not a standalone product, there's a boom in custom automations, citing an accessibility project that went from 55% accuracy after a year to 82% with ChatGPT 3.5 in a weekend. - www.jampa.dev

Do AI Website Builders Follow UX Best Practices? We Put Them To The Test - www.crazyegg.com

Editors Guild analyst Jason Hallock tested six AI movie script-analysis tools versus human readers, finding AI matched or beat humans on loglines but produced error-prone synopses and largely failed at nuanced analytical notes. - Variety

An apocalyptic vision of a million autonomous drones battling a million autonomous drones in the sky may sound like science fiction, but it’s coming. - Esquire

When Face Recognition Doesn’t Know Your Face Is a Face: An estimated 100 million people with facial differences are misrecognized or rejected by widely used facial-recognition systems, causing access failures for phone unlocks, IDs, airports and online services. - WIRED

The NBA is implementing automated officiating with arena cameras, sensors and machine learning to track the ball and players' body parts and adjudicate objective calls such as out-of-bounds and goaltending. (if they start to fairly call the 3-second rule and traveling, hell will break loose) - www.foxsports.com

Pranksters order 50 robotaxis to San Francisco's longest dead end street, triggering Waymo DDOS chaos - www.roadandtrack.com

I think we can retire the Will Smith spaghetti test? - Reddit

Follow the latest AI headlines via SkynetAndChill.com on Bluesky