AI Reading for Tuesday October 28

OpenAI said the US should commit to building 100 GW of new energy capacity per year, noting China added 429 GW last year versus 51 GW in the U.S. - CNBC

The US signed an $80 billion agreement to build nuclear reactors as part of an effort to support artificial intelligence development. - Bloomberg

Hyperscale cloud providers and developers are buying aeroderivative jet turbines, creating multi-year lead times into 2028–2030—Siemens Energy says over 60% of US turbine orders are AI-linked, Crusoe/OpenAI ordered 29 LM2500XPRESS units, about 1 GW. - Tom's Hardware

Surviving the AI Capex Boom: Big Tech goes from asset-light to capital-intensive - Sparkline Capital

OpenAI says about 0.07% of weekly active ChatGPT users show severe mental‑health symptoms and, after working with more than 170 experts, updated safety testing to cut substandard responses by 65–80% and achieve over 95% long‑conversation reliability. - OpenAI Help Center

A former OpenAI safety lead says the burden of proof is on OpenAI to show sexy chat is safe. A 2021 analysis found over 30% of player conversations in a text-based adventure using OpenAI models were explicitly lewd, prompting OpenAI to ban erotic use. - The New York Times

"I have major questions — informed by my four years at OpenAI and my independent research since leaving the company last year — about whether these mental health issues are actually fixed."

PayPal struck a deal with OpenAI to integrate its wallet into ChatGPT, prompting a roughly 13% premarket share jump and raising full-year adjusted EPS guidance to $5.35–$5.39 while reporting $458.1 billion in TPV and 7% FX-neutral growth. - Yahoo

Sen. Bernie Sanders called for breaking up OpenAI, warned AI could cause "massive" job losses, and proposed a "robot tax" to compensate displaced workers. - Futurism

Elon Musk's xAI launched Grokipedia on Monday, an AI-generated encyclopedia that promptly crashed at launch and contained over 800,000 entries compared with Wikipedia's nearly eight million, with sample entries including Elon Musk, OpenAI and President Trump. - The New York Times

xAI's Grokipedia says Grok fact-checked its entries and that some content is adapted from Wikipedia, and The Verge found some pages, such as the PS5 entry, are nearly identical to Wikipedia. - The Verge

Grokipedia is full of conservative AI slop - WIRED

Seems engineered to align with what Elon Musk would say - E. W. Niedermeyer

Makes up a lot of stuff - Kevin M. Kruse

but you can ask Grok AI to correct it, which is nice

Tesla chairman says Musk will walk if shareholders reject $1T pay package. Steve Jobs never did that LOL. - Quartz

A Tesla Optimus robot handed out candy near the Nasdaq building in Times Square while appearing plugged in, and Tesla has not confirmed whether the demonstration was autonomous or tele-operated. - Business Insider

Researchers at the Max Planck Institute found generative search engines cite less-popular deep-cut domains: Gemini's median cite was outside Tranco's top 1,000; 53% of Google's AI Overview sources weren't in the top 10 and 40% not in the top 100. - Ars Technica

Anthropic expanded Claude for Financial Services with a beta Claude for Excel sidebar, seven real-time market-data connectors, six finance-focused Agent Skills, and research-preview availability for Claude for Excel and Skills to paid Max, Enterprise and Teams tiers. - ZDNET

Claude for Excel launched as a beta research preview with a 1,000-seat waitlist for Max, Team, and Enterprise customers, providing workbook analysis, cell-level citations, formula preservation, error tracing and .xlsx/.xlsm support while excluding advanced Excel features. - Claude

Amazon will cut 14,000 corporate jobs—about 4% of roughly 350,000 corporate employees—SVP Beth Galetti announced, citing AI-driven transformation and CEO Andy Jassy's push to operate "like the world's largest startup," while offering severance and transition support. - Business Insider

Vercel trained AI agents by shadowing its top sales rep, automated inbound tasks, cut a 10-person team to one overseer, redeployed nine reps to outbound roles and plans to scale to hundreds of agents within 6–12 months. - Business Insider

At the 2025 Masters of Scale Summit, Box CEO Aaron Levie, LinkedIn's Aneesh Raman and Meta's Clara Shih said AI is reshaping work, warned of proliferating "workslop," urged context engineering and workflow reengineering to avoid only ~10% automation gains. - Fast Company

AI can stretch our minds—or stunt them - The Atlantic

Microsoft recently launched major Copilot updates—including group-chat support, a Real Talk mode, upgraded memory and the Mico avatar—while CEO Mustafa Suleyman said the company will never build a sex robot and urged peers to stop creating seemingly conscious AI. - MIT Technology Review

Apple Mac shipments up 14.9% as consumers favor Macs over Copilot+ AI PCs after Windows 10 support ends. - Tom's Hardware

Qualcomm announced AI200 and AI250 inference accelerator cards supporting 768 GB per card and near-memory computing, and preconfigured racks with direct liquid cooling and ~160 kW rack power, targeting datacenter inference. - The Register

Qualcomm unveiled AI200 (shipping 2026) and AI250 (planned 2027) inference chips, based on Hexagon NPUs, supporting up to 768 GB per card, deployable in liquid-cooled racks (~160 kW), with Humain as first customer; stock pops - CNBC

AI's memory problem, or context engineering to answer a prompt based on a set of facts - Geeky Gadgets

How much AI is the market willing to finance? - Financial Times

An estimated more than 10,000 AI startups exist worldwide, with over 2,000 receiving first-round funding last year as investors poured billions, creating a crowded, high-risk market likely to produce churn and later consolidation. - WIRED

24 AI Startups to Watch in 2026 - Bloomberg

Model makers, vibe coders, content creators, robot designers, office automators, killing machines

Wired AI issue: AI of a Thousand Faces - WIRED

Children increasingly interact with AI chatbots and apps, yet few ethical guidelines target them, creating risks like inappropriate content, biased recommendations and misuse; developers are adopting "Child Rights by Design", age filters, transparency and cross-sector oversight. - Hacker Noon

Alpha School promised AI learning, delivered a dystopia - WIRED

Waymo races accelerating Chinese competitors - Bloomberg

Follow the latest AI headlines via SkynetAndChill.com on Bluesky