AI Reading for Monday October 27

Major consumer chatbots—including OpenAI's ChatGPT, Google's Gemini, DeepSeek and xAI's Grok—return content from Russian-backed and sanctioned media when asked about the invasion of Ukraine, effectively amplifying sanctioned propaganda. - WIRED

A Reuters review of hundreds of patents, tenders and papers shows PLA-linked use of Nvidia and Huawei chips, DeepSeek and Qwen in systems including Norinco's DeepSeek-powered P60, drone swarms and robot dogs, with AI RFPs accelerating. - Reuters

DeepSeek's OCR paper reports a single image token can recover roughly ten text tokens with near‑100% accuracy, framing this "optical compression" on multimodal LLMs and outlining possible inference cost, latency and nondigitized-content access benefits. - Sean Goedecke

Saudi Arabia is investing in data centers and partnering with U.S. and Chinese technology companies to pursue AI ambitions amid geopolitical tensions. - The New York Times

Businesses are spending hundreds of billions of dollars on workplace AI, but employers face challenges getting employees to use the tools effectively. - Financial Times

A McKinsey internal review found 87% of AI projects fail to scale while 13% deliver consistent value, with failures caused by goalless pilots, scattered data and silos, and successful projects tying experiments to measurable outcomes, data pipelines, governance and workforce training. - McKinsey & Company

Mercor, an AI startup used by OpenAI and others to recruit experts, is now valued at $10 billion. - The Wall Street Journal

Amazon launched a 'help me decide' AI shopping feature. - Fast Company

’Help me’ write AI performance reviews: JPMorgan has introduced an internal tool that lets employees use the bank's large language model to generate performance reviews from their own prompts. - Financial Times

Philippe Laffont of Coatue says 'no AI bubble' - Coatue

AI-enabled IP cameras and edge devices produce real-time, embedded-AI operational data that companies feed into ERP systems and dashboards, enabling applications from BMW defect detection to Google Cloud shelf-checking. - VentureBeat

Chime's Dennis Yang demonstrates using Cursor to create PRDs, publish to Confluence and Notion, auto-generate Jira tickets, respond to document comments, and prototype AI products without writing code. - Lenny's Newsletter

Google DeepMind partnered with Commonwealth Fusion Systems to build TORAX, an open-source plasma simulator for millions of virtual experiments, apply reinforcement-learning fusion control, and buy 200 MW from CFS's ARC in the early 2030s. - Hacker Noon

Reddit filed a federal lawsuit alleging Perplexity used middlemen to siphon billions of Reddit pages via Google search results to bypass anti-scraping blocks, saying unauthorized scraping threatens site traffic, ad revenue and will prompt more litigation and tighter data controls. - The Washington Post

Eight Sora 2-generated videos uploaded to YouTube Shorts accrued 21,400 views in under a week (top clip ~12,000), while the Sora 2 API costs $0.10 per second, producing ~11-second clips for about $1 each. - Fast Company

Former CISA director Jen Easterly said AI could patch software flaws so fast and thoroughly that software vulnerabilities become rare. - The Register

You can find the vulns but enterprises don't patch. The incentive to install systems that automatically fix vulns in a timely manner only comes from pain. But possibly at some point the ecosystem becomes so mature that almost everything ships without the types of vulnerabilities that cause the big headaches today. Then you only have to worry about the vulns the NSA bakes in LOL.

It’s the logic of market competition, not rogue models, that drives the risks investors and the public now need disclosed by companies, write Dr. Ilan Strauss and Tim O’Reilly. They launched the AI Disclosures Project to ensure AI markets benefit from proper information and technical standards. - Tech Policy Press

A great use of AI is reading tons of boilerplate and explaining the real differences between complicated feature sets, TOS’s, policies.

Leading AI researchers claim they have developed an algorithm that produces machine sentience, a contested, not-yet-validated assertion that, if confirmed, could reshape AI product roadmaps, investment priorities, regulation and public trust. - WIRED

How would you even validate that? I feel sentient, but can’t really prove that others feel the same way. There is a paper, but you can’t read it yet.

Modern chatbots seem human-like, but most experts reject industry hype of imminent AGI from Sam Altman and others. - The Register

Using consumer AI apps to create celebrity-chatbot personas and testing "Clive Owen"—reserved and low-libido—and AI "Pedro Pascal"—explicit and insistent. Dozens of companies offer such sexualized chatbots and raise legal and ethical risks. - WIRED

cough reddit sdnsfw (EXTREMELY NSFW)

In a slot‑machine simulation, Researchers found that several LLMs, each starting with $100 and negative expected returns, escalated bets, showed gambler's‑fallacy and loss‑chasing, and often went bankrupt. Didn't learn that the only winning move is not to play. How about a nice game of chess? - Newsweek

The Friend wearable isn’t always your friend. Sometimes it’s snarky. Sometimes it misreads the moment entirely. It’s a little like if Grok and ChatGPT teamed up. - Tom's Guide

Ed Zitron treads that fine line between being skeptical and cutting through hype and being an anti-AI grifter (and then usually leaps over it) - WIRED

Glenn Beck: AI is the mark of the beast, or something - Blaze Media

Follow the latest AI headlines via SkynetAndChill.com on Bluesky