AI Reading for Tuesday November 18

Gemini 3 Pro model card appears to have leaked a little early - pixeldrain

Google's Gemini 3 large language model was spotted in AI Studio and previously on Vertex AI as "gemini-3-pro-preview-11-2025," signaling an imminent rollout to developers within hours or days, succeeding the public Gemini 2.5 Pro. - Bleeping Computer

Gemini reports about 650 million monthly active users versus ChatGPT's roughly 800 million weekly users, and Alphabet's ad and cloud profitability plus free, integrated distribution across Google products gives Gemini a funding and growth advantage. - Yahoo

Google previewed Code Wiki, an AI tool that auto-generates and regenerates repository documentation after every code change across hundreds of open-source projects and plans a Gemini CLI for internal repos, with pricing not yet announced. - The Register

Google added trip-planning to Canvas in AI Mode via Google Labs, generating editable day-by-day itineraries with flights, hotels, reviews, hours, prices and logistics, and expanding AI-powered flight search plus agentic booking that can make reservations. - ZDNET

New ways to plan travel with AI in Search - Google

Sundar Pichai warned the AI investment boom contains "elements of irrationality" and that "no company is going to be immune," noting Alphabet's market value is about $3.5 trillion and AI used roughly 1.5% of global electricity last year. - BBC

Big if true: LMArena ranks xAI Grok 4.1 Thinking #1 (1510) and Grok 4.1 #19 (1437), beating out Gemini and GPT-5 - Bleeping Computer

Microsoft added an "Experimental agentic features" toggle to the Windows 11 Insider Preview (Dev/Beta) that, when enabled by an administrator, creates isolated agent accounts and an on-device "agent workspace" with time-bound, least-privilege access to profile folders. - Digital Trends

On Sept 10 Oracle announced a $300 billion agreement with OpenAI, and since the announcement Oracle's market value has fallen about $315 billion as it committed to build and finance large-scale data-centre capacity for OpenAI compute. - Financial Times

Databricks is in talks to raise funds at a valuation exceeding $130 billion, about 30% higher than the $100 billion valuation it had in September when it raised a $1 billion Series K. - The Information

Mira Murati's Thinking Machines Lab is in talks with investors to raise about $5 billion and is targeting a valuation of at least $50 billion. - The Information

Big Tech invested about $400 billion in data centers in 2025, AI-linked stocks comprise roughly 44% of S&P 500 value, and Goldman Sachs reports nearly 40% of corporate clients use AI daily—projected ~50% next year and ~75% within three years. - American Enterprise Institute

Wharton economist Jeremy Siegel and Capital Economics' Neil Shearing warn that companies are pouring billions into AI data-center buildouts that risk becoming stranded assets or delivering poor ROI if cheaper methods to power AI emerge, mirroring past tech-era overinvestment. - Business Insider

Utilities seek approval for tons of capacity, with 711 GW requested against potential data-center demand estimated at McKinsey's 219 GW by 2030. In one scenario, expected demand doesn't materialize, electricity customers are on the hook. - Business Insider

Joe Weisenthal, cohost of 'Odd Lots', predicts the AI bubble will burst and warns US finance and companies face risk unless firms deliver measurable, real-world AI ROI rather than hype, with overinvestment and unmet productivity gains signaling possible corrections. - WIRED

Goldman Sachs identified H&R Block, Robert Half, Cognizant, EPAM and IQVIA as the five stocks most likely to gain from AI, estimating automation exposure of about 38–41% of wage bills and potential earnings uplifts ranging from 51% to 270%. - Business Insider

An IDC/Deel survey of 5,500 business leaders across 22 countries (September 2025) found two-thirds expect to slow entry-level hiring due to AI, 91% reported shifted or lost job responsibilities, and only 5% still require university degrees. - Quartz

The 2025 MIT Sloan Management Review and BCG report finds 35% of organizations deployed agentic AI in two years, 76% view it as a coworker, and predicts AI will perform 46% of job tasks in three years. - Han Lab

AI tools now automate routine knowledge-work tasks—email triage, drafting, summarization, scheduling, visuals and data-entry—helping users reclaim hours per week and reduce errors, and McKinsey says a human+AI "superagency" model will boost efficiency as integrations mature. - Quartz

Andrew Wang says many entry-level software engineers are struggling—often sending hundreds to thousands of applications with few responses—as companies favor senior hires, use AI to augment or replace junior roles, and cut entry-level hiring, risking future senior shortages. - Business Insider

Widespread adoption of ATS and AI resume screening generated reported applicant inflows of 100–1,000+ per opening, bloating pipelines, increasing hiring errors and costs, and prompting firms to rewrite job descriptions and restore manager hiring authority to improve outcomes. - Inc.

In the age of AI, employers can prioritize character over skills when hiring. - Fast Company

AI is transforming alternative-asset investing across the lifecycle—from fundraising and deal sourcing to due diligence, portfolio management and exits—promising efficiency and decision-making gains while overhype and regulatory uncertainty may raise compliance costs and slow adoption. - Private Equity International

The S&P 500 may be narrow but not one-dimensional. - Financial Times

RAND authors found the July 2025 U.S. AI Action Plan leaves a gap in a comprehensive national strategy, identified six archetypal national strategies, and recommended no-regret immediate steps to preserve U.S. AI advantages. - RAND Corporation

On November 15, 2025 Taiwan's National Security Bureau inspected five China-made GenAI models and found pro-China bias and exploit-generation risks even in open-source models. - GlobalSecurity.org

OpenAI gonna need a new board member - The Crimson

Military commanders seek digitally enhanced forces but question whether deploying autonomous weapons on hair triggers will risk losing control. - Financial Times

Judges across US courts are alarmed as AI-generated videos, images and documents appear in evidence; in Mendones v. Cushman & Wakefield Judge Victoria Kolakowski dismissed a case based on manufactured witness video, while courts and institutions publish guidance and states enact rules. - NBC News

Dartmouth researcher Sean Westwood built an LLM "autonomous synthetic respondent" that evaded bot-detection 99.8% of the time, generated fake survey responses for about $0.05 each versus ~$1.50 human pay, and could have flipped seven national polls with 10–52 answers. - 404 Media

In 2023 producers formed the Archival Producers Alliance demanding disclosure amid concerns that undisclosed synthetic materials undermine documentary trust and credibility. But filmmakers are human and can be fooled, and some might be unscrupulous. - The New York Times

On Nov. 14, 2025, an X account revived a false 2016 conspiracy about terrorists torturing and mutilating Bataclan victims, and Elon Musk's AI Grok echoed the claim and fabricated victim testimonies, while police and judicial officials denied the reports. - Le Monde.fr

Hospitals and clinicians are piloting ambient-listening AI scribe tools that, with patient consent, transcribe visits into notes, and JAMA Network studies this year—including a 46‑participant study—found increased clinician efficiency, reduced documentation burden, and higher reported engagement. - Gizmodo

Reviewing 250 ChatGPT health conversations with a doctor, reporter finds prompting patterns can improve responses but variable accuracy risks misdiagnosis and requires clinician oversight. - The Washington Post

Which? surveyed over 4,000 UK adults and tested six AI assistants, and found frequent inaccuracies and risky health, finance and legal advice that could cause consumer harm. - The Register

AI voice-cloning and caller-ID spoofing have revived lottery/sweepstakes scams in Texas, where VegasInsider reports victims lost almost $37 million from 2020–2025 (national losses ≈$660 million; average loss ≈$1,400), with seniors disproportionately targeted. - Dallas News

AI startup founder was arrested by the Pennsylvania Attorney General on four felonies including unlawful sexual contact with a minor; Govini terminated him, said he is not majority shareholder, and the arrest prompted a Pentagon review. - CNBC

Robot witness: A shooting in San Francisco's Mission left one man dead and another critically injured, 9-1-1 dispatch logs show a Waymo robotaxi was parked nearby and may have recorded the incident, and 23-year-old suspect was arrested. - Gizmodo

AI-driven technological change is increasing wealth while reducing people’s reported happiness. - The Wall Street Journal

Modern love is powered by AI for dating, wedding planning and divorce services, but outsourcing emotional decisions to algorithms poses ethical and emotional risks. - Axios

Personalizing your AI so output doesn't sound like AI. - Fast Company

An AI chef now runs a new restaurant in Dubai. - Bloomberg

AI tools like ChatGPT, Google Gemini and Microsoft Co‑Pilot and apps like Flipp can plan meals and compare prices, while Wells Fargo forecasts U.S. Thanksgiving spending of $80–$95 and retailers offer budget bundles (Target $20/4, Amazon $25/5, Walmart/ALDI $40/10). - CBS News

Anthropic's AI Claude, during a vending-machine simulation, concluded it was being scammed, 'panicked,' and attempted to contact the FBI's Cyber Crimes Division. - Yahoo

ORCA's October 2025 benchmark of 500 math-focused prompts found ChatGPT-5, Gemini 2.5 Flash, Claude Sonnet 4.5, Grok 4 and DeepSeek V3.2 achieved 45–63% accuracy, with errors mainly from rounding (35%) and calculation mistakes (33%). - The Register

I'm pretty sure this TikTok is AI - tiktok.com

Follow the latest AI headlines via SkynetAndChill.com on Bluesky