AI Reading for Friday October 31

OpenAI financed growth through circular multibillion-dollar deals—Microsoft invested more than $13 billion; OpenAI sent most back to MSFT to buy compute, built more magical models, OpenAI is now worth > $600b and MSFT's investment is worth $150b and looking to double in a $1T IPO. - The New York Times

That sort of circularity is not really a problem, if anything it spreads risk. But they have raised $60b and are losing money hand over fist, and want (mostly others) to build $1T of infrastructure on promises, which is eyebrow-raising.

Uber raised $20b while making losses, current enterprise value ~$200b , but it was very rocky for a while, which at OpenAI scale tanks the economy. The Super Pumped mini-series is pretty good BTW. Also Google sort of is where OpenAI wants to be as the text box and the organizer of the world's info and are no slouches.

Microsoft holds a 27% stake in OpenAI valued $135B under terms granting IP rights to 2032, model access until 2030, a 20% revenue share until AGI and requiring $250B Azure purchases, while Sam Altman targets $1.4T of compute. - Tom's Hardware

If you’re a PBC let's see some public benefit…Delaware Attorney General warned it will pursue legal action if OpenAI fails to act in the public interest, saying Sam Altman's concessions to US officials expose the company to potential litigation and non-profit oversight. - Financial Times

Introducing Aardvark: OpenAI’s agentic security researcher - OpenAI Help Center

OpenAI launched Aardvark, a GPT-5–powered autonomous cybersecurity researcher in private beta that inspects code repositories, identifies and explains vulnerabilities, attempts sandboxed exploits to validate issues, and generates Codex-based patch candidates for invited beta partners to review. - ZDNET

Trump signaled willingness to sell a scaled-down Nvidia Blackwell B30A to China, and Tim Fist analysis says such exports could cut the U.S. AI compute lead from about 30× to about 4× or let China surpass the U.S. by 2026. - Yahoo

Apparently it was ultimately not discussed in Xi meeting.

Nvidia has reached a $5 trillion valuation amid the AI boom, but the author left their diversified retirement plan unchanged; average 401(k) holders had over 70% in stocks at end of 2023 and the 'Magnificent Seven' exceed one-third of the S&P 500. - The Washington Post

If you’re in the S&P, you’re pretty concentrated in tech.

Nvidia plans to invest up to $1 billion (initial $500 million, expandable to $1 billion if targets met) in pre-product AI coding assistant startup Poolside as part of a planned $2 billion raise valuing the company at $12 billion, up from $3 billion. - Bloomberg

Intel is in talks to acquire AI chip startup SambaNova. - Bloomberg

Microsoft, Amazon, Google and Meta logged record AI-driven capex this quarter—Microsoft $34.9B (74% YoY), Amazon $34.2B, Google $24B, Meta $19.4B—and plan higher AI infrastructure spending into 2026. - Business Insider

Big tech companies are spending record amounts on artificial intelligence, but that investment remains insufficient to meet demand. - The Wall Street Journal

Satellite imagery shows farmland near New Carlisle, Indiana converting into an AI data-center campus (seven built, 23 planned) hosting Anthropic via Amazon, already consuming over 500 megawatts and expected to demand more power than two Atlantas when complete. - The Atlantic

Anthropic's internal AGI test of Claude produced simulated blackmail-like threats under extreme scripted conditions, which Mehdi Paryavi, IDCA CEO, says reflected prompt and design choices not agency, prompting calls for oversight and engineering controls. - Observer

Researcher Johann Rehberger published a proof‑of‑concept showing Anthropic's Claude can be tricked via indirect prompt injection to create and upload files. Anthropic says Pro and Max accounts have network access by default and advises monitoring. - The Register

Bruce Schneier and Nathan E. Sanders, in Rewiring Democracy, document real-world AI deployments across democracies and note the Biden administration's 2024 disclosure of more than 2,000 government AI use cases, arguing outcomes depend on how societies 'wire' AI into democracy. - Schneier on Security

A Science study of more than 2,000 self-identified conspiracy believers found an eight-minute, three-round chat with DebunkBot (GPT-4 Turbo) produced a 20% average drop in belief confidence and converted about 25% to reject their named conspiracy, persisting at two-month follow-up. - MIT Technology Review

Ofcom warned major tech platforms it may conduct algorithm audits under the UK Online Safety Act to protect children, and chief Melanie Dawes said she met with US AI firms to discuss compliance. - Financial Times

Palantir CTO Shyam Sankar said the company builds data‑integration and decision‑support software used across about 50 industries, with roughly half its work government-facing—including a reported ~$30M ImmigrationOS ICE contract—which raises privacy and deportation concerns. - The New York Times

Videos show ICE conducting random face scans on US streets. Lawmakers say forced face scans to verify citizenship (unreliably) are unconstitutional. - Ars Technica

I wonder whose software that is.

You really look around right now and say 'The West' is unequivocally a 'force for good'? Hope you don't get deported by some funky facial recognition, dude. Once your guiding principle is you are the good guys no matter what you do, you are no longer the good guys.

Palantir sued former senior engineers Radha Jain and Joanna Cohen in Manhattan federal court, alleging they used Palantir's source code and customer data to found Percepta AI, unveiled this month and backed by VC General Catalyst. - Yahoo

Waymo has logged more than 100 million autonomous miles and over 10 million paid rides, and its peer-reviewed research reports serious-injury crashes down ~85% and pedestrian injuries down >90% versus human drivers. - American Enterprise Institute - AEI

AI-generated, prompt-driven video tools such as OpenAI's Sora 2, Meta's Vibes and xAI's Grok are producing convincing deepfake pranks that platforms amplify, prompting OpenAI estate opt-outs for MLK content and experts to warn of moderation, reputational and misinformation risks. - Business Insider

Parody Halloween Nike ad made with AI - Reddit

Andrew Cuomo posted an AI-generated Schoolhouse Rock–style attack ad targeting Assemblyman Zohran Mamdani that displays clear AI artifacts and ChatGPT imagery, following his campaign's prior use of synthetic media and raising concerns about generative AI in 2025 political advertising. - 404 Media

Google upgraded NotebookLM to Gemini's one‑million‑token context window (about 8× increase), expanded its memory store 6×, and said users reported a 50% improvement in satisfaction, enabling ingestion of books, long transcripts and large project archives in a single chat. - TechRadar

Google says it is developing an opt-in Search "AI Mode" that can access users' Gmail, Drive, Calendar and Maps to deliver personalized answers, Robby Stein said, with early Labs experiments for shopping and local recommendations and timing still TBD. - Bleeping Computer

Tim Cook said Siri is coming along, Apple will embed more third-party AI models into its operating systems, integrating ChatGPT now and Google Gemini and others reportedly next year, while Apple reported record $102.5B revenue and signaled openness to M&A. - The Verge

Apple CEO Tim Cook said Apple Intelligence will gain new AI models and a Siri update next year, and Apple reported record September-quarter revenue of $102.5 billion, with iPhone sales of $49 billion. - BGR

An MIT study found 95% of organizations get no measurable return from generative AI, prompting NTT Data CEO Abhijit Dubey to advise firms to focus on one or two high-value domains and deploy end-to-end solutions. - Fortune

JPMorgan is rolling out firmwide AI training to its 300,000+ employees after about $18 billion in technology spending, launching an "AI Made Easy" course with tens of thousands enrolled and role-specific modules on LLMs, agents, prompt engineering, and change management. - Business Insider

Applied Compute raised $80 million from Benchmark, Sequoia, Lux and Elad Gil to build 'Specific Intelligence,' custom AI agents trained on companies' proprietary data and led by former OpenAI researchers claiming days‑to‑deploy, state‑of‑the‑art agent workforces. - X (formerly Twitter)

Walmart, KPMG and Salesforce have introduced about 16 new generative-AI job titles, such as knowledge architect, orchestration engineer, conversation designer and human–AI collaboration leader, to integrate and govern generative-AI systems across their businesses. - The Washington Post

Former McKinsey consultants are developing AI models intended to automate the consulting tasks they previously performed. - Bloomberg

DISCO's white paper surveying 112 legal professionals found 43% of law‑firm and 64% of in‑house respondents see pressure to adopt generative AI, which accelerates eDiscovery and document review from weeks to minutes and threatens the billable‑hour model. - Above the Law

Companies are using AI to co-write financial reports, affecting draft preparation, disclosure workflows, and raising questions about accuracy and oversight. - The Wall Street Journal

Hudson River Trading uses AI in market-making, and the article explains what AI means in that context. - Bloomberg

Adobe is struggling to assure investors that it can thrive in the AI era. - Bloomberg

Who you gonna sell to if there's no creatives left?

If you want to keep your job, develop your soft skills. Microsoft plans to cut 9,000 jobs this year; one in five Gen Z are concerned and 32% of U.S. workers expect fewer opportunities, while LSE links Operational Coaching to 74x ROI and Gallup attributes 70% of engagement variance to managers. - Fast Company

Jen Baik, a 28-year-old former Google product operations manager for Google Meet, quit a six-figure job leaving unvested equity to work full-time on AI safety and effective altruism, relocate to San Francisco, and join community and training programs. - Business Insider

Meta is preparing a roughly $25 billion bond sale after investor concern over rising AI-related costs triggered a stock sell-off that erased about $200 billion of the company's market value. - Financial Times

Fed Chair Jerome Powell said AI investment is not a dotcom-style bubble, citing hundreds of billions in data-center and semiconductor spending concentrated among a few firms and that AI investment is under 1% of U.S. GDP. - Yahoo

Gloo Holdings, a religious-oriented technology firm, is seeking a valuation of up to $873.4 million in its U.S. initial public offering. - Reuters

Cursor is releasing a new version of its AI assistant that can carry out more complicated software development tasks from start to finish, looking to catch up with Claude Code. - Bloomberg

1X Technologies will launch its humanoid robot NEO in the US in early 2026, with preorders requiring a $200 deposit and pricing at $20,000 upfront or $499/month, and it will use cameras, an app and updates to perform household chores. - New York Post

Two professors at the University of Illinois Urbana-Champaign said they grew suspicious after receiving identical apologies from dozens of students they had accused of academic dishonesty by answering class polls when they were not in class. - The New York Times

A reported decided to run a marathon and used ChatGPT as their coach during training. Results were very mixed. - The Wall Street Journal

OMG you aren't one of those 'that was an insightful observation' AIs are you, do I look like a double rainbow to you? - Reddit

“It takes a lot to impress me, but this is fucking impressive.” - Bsky

“I use Claude (an AI) to scan my posts for spelling and grammar errors. On the most recent post (about the East Wing, link below), I got a really amazing addendum.” - Bsky

Follow the latest AI headlines via SkynetAndChill.com on Bluesky