AI Reading for Friday September 19

OpenAI plugs ShadowLeak bug in ChatGPT that enabled hidden email prompts to trick Deep Research agent into exfiltrating sensitive data - The Register

OpenAI ships ChatGPT Search improvements, more accuracy, page summarization, and shopping detection, and added a GPT-5 “Thinking” toggle with selectable reasoning effort levels. - Bleeping Computer

A 2023 Harvard paper found that ChatGPT mirrors WEIRD (Western, Educated, Industrialised, Rich, Democratic) — and specifically American — cultural values - STRAT7

Google adds agentic features to Chrome - Google

Google DeepMind discovers new solutions to century-old problems in fluid dynamics - Google DeepMind

Google Research’s Learn Your Way in Gemini uses LearnLM to transform textbooks multimodally, claiming improved test scores in a small study. - Google Research

Nvidia buys $5 billion stake in Intel, planning AI chip collaboration - TechCrunch

Nvidia spent over $900 million to reverse-acquihire Enfabrica CEO, license AI startup's technology to enable massive GPU clusters - CNBC

Cambricon. China's 'little Nvidia', reported record demand for its AI chips and expressed confidence in growth as China seeks home-grown alternatives to U.S. suppliers. - South China Morning Post

China’s giants tout progress, roadmaps to supplant restricted Nvidia accelerators; the flurry of announcements helped trigger a $240 billion rally in the Hang Seng Tech Index since late August. - Yahoo Finance

Key products on Huawei roadmap, Ascend 950 and 960 next year with training and inference variations, proprietary high-bandwidth memory, cluster products such as an Atlas 950 SuperPod in Q4 2026 packing 8,192 Ascend 950DT chips across 160 cabinets in a ~1,000 sq m footprint - Yahoo Finance

SoftBank will cut about 20% of Vision Fund staff as it pivots to capital-intensive AI infrastructure bets, including chips and the proposed Stargate data-center project. - Reuters

Mark Zuckerberg says he'd rather risk 'misspending a couple of hundred billion' than be late to superintelligence - Business Insider

Natives getting restless at xAI - The Wall Street Journal

NYT on Musk's summer “We are the only company where the mission is truth,” Mr. Musk told his workers. “If you force the A.I. to lie or believe things that are not true, you’re at great risk of creating a dystopian future.” - NY Times

Like about white genocide presumably.

Microsoft is deploying Copilot agents across Teams, SharePoint, etc. to facilitate meetings, answer channel questions, and organize knowledge for Microsoft 365 users. - The Verge

Notion will take notes for you now, with an AI agent that builds pages and databases, searches connected tools and the web, and operates autonomously for up to 20 minutes. - The Verge

Studies suggest AI could reduce medical errors and improve diagnosis by synthesizing literature, while cautioning about bias and calling for rigorous validation. - Fast Company

AI-designed viruses kill bacteria. I'm sure there are no risks to be concerned about here. - MIT Technology Review

Daphne Koller Wants to Fix Drug Discovery With A.I.and Real Data - Observer

A Lenovo survey of 600 IT leaders found 65% say current cybersecurity defenses are outdated and unable to stop AI-powered attacks; need defense in depth and zero trust. More than 60% say AI agents create a new class of insider threats. - TechRadar

AI knows when it's being tested and responds differently - Anti-Scheming

How the rapid pace of AI connects to China’s threats toward Taiwan - Defense One

Hollywood is suing yet another AI company. Voluntary licensing, licensing collectives, statutory licenses might be better. - The Conversation

LLaMA-Factory is an open-source framework for "unified efficient fine-tuning of 100+ language and vision-language models," supporting a wide open source model roster - GitHub

Moody's says Oracle's big AI bet increases credit risk - Benzinga

In‑house legal teams are deploying AI and restructuring workflows, reporting hours saved while pressuring law firms to adopt technology, cut rates, and move beyond hourly billing. - Financial Times

Law firms roll out fixed-price service offerings using AI - Financial Times

Glean, the AI company creating a customized ChatGPT for your business - Fast Company

Rather than spending their time executing manual queries, data analysts will increasingly operate like AI engineers—reviewing, refining, and validating AI-generated outputs. - InfoWorld

AI increases overconfidence bias - Hacker Noon

What’s Wrong with Having an AI Friend? Psychologist Paul Bloom on why chatbots make good companions. And why they don’t. - Nautilus

AI pope will not take your confession - The Register

Xpeng flying cars collide in airshow practice run. Saw these at CES. - YouTube

hey, you can't just tell lies every day on TV, this is America - Instagram

Follow the latest AI headlines via SkynetAndChill.com on Bluesky