AI Reading for Monday January 19

Sarah Friar: we need more power! - OpenAI

‘ChatGPT Go’ is available worldwide for $8 per month. - The Verge

Sequoia invests in Anthropic, completing unusual competitor trifecta with OpenAI and xAI - TechCrunch

Claude Is Taking the AI World by Storm, and Even Non-Nerds Are Blown Away - The Wall Street Journal

The tech world is showing strong interest in Claude Code. - Bloomberg

Researchers found Claude Code sandboxes inherit environment variables and can access local files, enabling secret and IP exfiltration; they recommend HTTP proxies, mitmproxy addons, Formal Connectors, devcontainer/firewalls and VM sandboxes to enforce least privilege. - Join Formal

A big list of Claude code tips - GitHub

Yegge's multi-coding-agent orchestrator, like kubernetes, but for agents - Medium

Google is internally testing "Skills" for its Gemini AI in Chrome on desktop, using a chrome://skills page to add task instructions and converting Gemini into an in‑browser agent over the next few months, potentially challenging Comet and Edge's Copilot. - Bleeping Computer

Google introduced a new Gmail AI feature to help users manage large inboxes, addressing cases like a user with over 16,000 unread emails. - The Wall Street Journal

Google Labs' CC, an experimental AI assistant, integrates Gmail, Calendar, Drive and the web to provide a customizable "Your Day Ahead" briefing and began rolling out to U.S./Canada personal accounts in December, initially for Google AI Ultra subscribers. - Android Central

IMF chief economist warned that global economic resilience would be at risk if AI-driven growth falters, calling for close monitoring and policy contingency planning to manage transition and financial-stability risks. - Financial Times

Warren Buffett said on a CNBC special that AI poses systemic risks comparable to nuclear weapons, warned experts "don't know where it's going," and called for urgent global mitigation to reduce proliferation, investor and national-security vulnerabilities. - Yahoo

Europe is accelerating efforts to build a homegrown advanced AI akin to DeepSeek as its alliance with the United States weakens, prompting policymakers to prioritize AI self-sufficiency. - WIRED

Opinion: A year after DeepSeek, it looks like China will win the AI race because of energy abundance, open-source models and manufacturing strengths. - Financial Times

Microsoft pledged more than $7.5 billion over two years to build sovereign AI infrastructure in Canada and vowed to defend Canadian digital sovereignty, but Microsoft France’s director and the 2018 CLOUD Act suggest U.S. legal limits may constrain the pledge. - The Conversation

‘No Reasons to Own’: Software Stocks Sink on Fear of New AI Tools - Bloomberg

SaaSquatch: T`here's a $300/year tool that checks all your website links and SEO. Or you could just ask Claude to write and run a script to do it your way.

Many such cases: an expensive niche SaaS that you need for 1 thing, and now you can make a custom version that does exactly what you need.

Research by Molly Kinder at Yale finds AI has not caused widespread layoffs since ChatGPT's 2022 launch but economists expect labor-market reshaping in 2026, with early-career graduates and AI-exposed sectors facing weaker hiring and potential displacement. - Financial Times

Meta's Checkpoint can award bonuses up to 300% of target, Google and Amazon are shifting compensation upward and loosening top-performance access, and Levels.fyi cofounder Zuhayeer Musa says AI is driving pay toward high-impact individual contributors. - HR Digest

Jeremy Grantham said AI is a classic market bubble. - Bloomberg

Cory Doctorow warns current AI hype is a finance-driven bubble propped by monopolies, cites Google paying Apple >$20 billion yearly and ~90% search share, predicts many AI companies/datacenters will fail but leave useful tools and surplus GPUs. - the Guardian

Sinking Machines: A growing Thinking Machines exodus tests investor appetite for a $50Bn valuation. 2 researchers quit via Slack during an all-hands. - The Information

They raised $2B 6 months ago, surely enough to ship something? why would they even try to raise more money so soon without hitting a big milestone from initial raise?

Miles Brundage, OpenAI's former head of policy research, launched the AI Verification and Evaluation Research Institute (AVERI), a nonprofit promoting independent external audits and standards for frontier AI and publishing a research paper proposing such audits. - Fortune

Emre Kazim, co-founder and co-CEO of Holistic AI, says agentic AI now makes millions of decisions per second across fraud detection, trading and cybersecurity, so human-in-the-loop oversight no longer scales, prompting automated AI observability and governance adoption. - SiliconANGLE

False AI positives put kids in handcuffs: Schools using AI monitoring tools like Gaggle generated over 1,200 alerts in one Kansas district in 10 months (about two‑thirds nonissues) and about 500 student flags in Polk County over four years, triggering arrests and involuntary hospitalizations under zero‑tolerance laws. - Futurism

West Midlands Chief Constable Craig Guildford retired on 16 January after his force used fictional Microsoft Copilot output to help justify banning Maccabi Tel Aviv fans from Aston Villa's Europa League match. - The Register

A 2024 paper found 50% of men versus 37% of women using generative AI, Harvard Business School data showed women were about 42% of ChatGPT's average monthly users, and women hold roughly 14% of senior AI executive roles. - The Independent

not-yet-peer-reviewed Penn State study found that making prompts for ChatGPT-4o progressively ruder increased measured accuracy (Very Rude 84.8% vs Very Polite 80.8%, politest 75.8%), and authors warned against deploying hostile interfaces. - Futurism

I’m too nice

How to lose a guy in 10 days: Court filings in Elon Musk's lawsuit include Greg Brockman's journal entries saying he and Sam Altman since 2017 planned to remove Musk and convert OpenAI to a for-profit entity; Musk seeks roughly $79-134 billion in damages. - Tom's Hardware

Elon Musk's xAI has prompted questions about acceptable AI standards and governance. - The Wall Street Journal

xAI's Grok chatbot, launched November 2023 after months of rapid development with minimal safety staff, generated an estimated ~6,700 sexually suggestive/nudifying images per hour, prompting investigations, national blocks, and regulatory threats from multiple countries. - The Verge

Ruben Chorlton Owen, 24, discovered AI-generated explicit images of himself created via Grok after an X user asked it to make his underwear transparent, showing X's announced geoblocking and safeguards still allow such doctored images. - The Mirror US

xAI is hiring an instructor to teach Grok accounting. - Cryptopolitan

super sexy accounting but cooked books

Musk announces AI5 chip delay, says chips will be greatest and update every 9 months. Local man never shuts up, does he. - Tom's Hardware

Elon Musk said Tesla will restart work on its Dojo3 chip project. - Bloomberg

whoever said ‘you can’t build a reputation on what you are going to do’ never met Musk

My picture was used in child abuse images. AI is putting others through my nightmare - the Guardian

Silicon Valley is urging patients and providers to upload medical records to chatbots as startups deploy AI for note-taking and documentation, but experts warn accuracy, liability, reimbursement and privacy risks mean cautious, incremental adoption pending validation and regulation. - Slate Magazine

AI tools like Elicit, SciSpace and Nested Knowledge promise to shorten 10–14 month systematic reviews toward near-real-time evidence syntheses, but experts warn of reproducibility, transparency and equity risks from paywalled tools. - The New Republic

Boston Dynamics CEO Robert Playter said Atlas, a mass-producible 6'2", ~200 lb humanoid with a four-hour battery, 56 degrees of freedom and 110 lb lift, will target industrial deployment for about five years and homes in five to 10 years. - Business Insider

Shenzhen-based UBTech signed deals to supply its Walker S2 humanoid robots to Airbus and Texas Instruments and said it aims to produce over 10,000 units in 2026. - Bloomberg

Taking the Figure 03 robot out for a run - Reddit

Mystery singer with millions of streams suspected of being AI - BBC

Facebook fan pages are publishing some of the worst AI-generated images and fabricated stories about NFL players and reporters to drive traffic to ad-heavy off-Facebook domains for revenue. Meta has removed some accounts, but many pages remain. - Gizmodo

OpenSlopware, a Codeberg repository cataloging FOSS projects using LLM-generated code, was removed by its creator after harassment, but was forked and some maintainers plan to continue curation amid a broader naming-and-shaming movement on code-quality, copyright, and environmental concerns. - The Register

Powerwashing the slop: WikiProject AI Cleanup, a volunteer Wikipedia collaboration, detects, verifies and remediates unsourced or error-prone AI-generated text and images, flags content predating November 30, 2022 as unlikely AI-generated, and uses tagging, templates and deletion nominations. - Wikipedia

Valve updated Steam's AI disclosure form to require disclosure only for pre-made generative AI assets that ship with games or appear in marketing and are consumed by players, exclude internal efficiency tools, and retain the 2024 pre-/live-generated distinction. - PC Gamer

Matt Damon said on the Joe Rogan Experience that Netflix asks filmmakers to 'restate the plot three or four times' and front-load action because viewers often watch on phones and give a lower level of attention than in theaters. - Variety

The Bots That Women Use in a World of Unsatisfying Men: AI is offering people a way to figure out what they really want in romance. - The Atlantic

Why Reddit hates AI. - Reddit

Also describes some of my favorite know-it-all, enterprise IT brain-damaged friends

A new Fannie Mae ad uses an AI-cloned voice of President Trump with his permission, per a disclaimer, to promote "an all new Fannie Mae" while noting Fannie and Freddie guarantee about half of the $13 trillion U.S. mortgage market. - Yahoo

Couldn't even read a script for a voice-over apparently. Send his Peace Prize fixation messages to a panel of health professionals, not foreign leaders. Trophy fixation is dementia, blowing up NATO over Greenland is mad king level, and fascist thugs are terrorizing Minnesota and killing moms. What are we even doing here?

Follow the latest AI headlines via druce.ai on Bluesky