AI Reading for Monday November 24

The U.S. economy has become increasingly reliant on rising AI-related investment by businesses and government. - The Wall Street Journal

Big technology companies are increasing debt to finance artificial intelligence projects, a trend that could strain credit markets. - Bloomberg

A surge in bond issuance tied to artificial intelligence projects is adding pressure to financial markets. - The Wall Street Journal

Major tech firms sold about $90 billion of bonds to finance AI-ready data centers, and J.P. Morgan estimates up to $1.5 trillion over five years—potentially over 20% of the U.S. investment-grade market by 2030—prompting caution from DoubleLine. - Yahoo

AWS's Bedrock faced 'critical capacity constraints' over the summer, causing tens of millions in lost or delayed revenue—Epic ragequit a $10M Fortnite project and customers faced $52.6M in projected delays—and Amazon is adding over 3.8 GW of power. - Business Insider

Documents show Amazon's count of data centers has topped 900 amid increased AI investment. - Bloomberg

Meta opened a large new data center that highlights the intersection of AI workloads and contested accounting practices. - The Wall Street Journal

Nvidia's earnings are not evidence of an AI bubble, and the company's margins should remain safe if power constraints, rather than chip scarcity, limit deployment. - Stratechery by Ben Thompson

Alibaba's new Qwen AI app debuted strongly and is positioned to compete with OpenAI's ChatGPT. - Bloomberg

Salesforce CEO Marc Benioff said he switched from OpenAI's ChatGPT to Google's Gemini 3 after two hours' testing, calling it a major leap in reasoning, images and video, and other executives echoed praise, intensifying competition with OpenAI and Anthropic. - Business Insider

OpenAI and Jony Ive have strip-mined more than 40 former Apple hardware engineers in recent months, including leaders from industrial design and hardware, affecting camera, silicon, testing, audio, wearables and potentially slowing work on Siri and other projects. - BGR

A research leader who led OpenAI's model policy team, which oversees AI safety research and ChatGPT's crisis-response and mental-health interactions, is leaving the company. - WIRED

probably fine, just needs a mental health break. is there a popup that says, you’ve been working at OpenAI for a while, is this a good time to take a break?

Research finds closed models dominate ~80% of monthly LLM tokens and cost ~6× more than open models despite modest performance gains, and shifting demand to open models could cut prices >70% and save about $24.8 billion in 2025. - Reddit

But OpenAI is also losing money on inference 🤔

Andrej Karpathy published llm-council on GitHub, a local web app that uses OpenRouter to query multiple LLMs (e.g., GPT-5.1, Gemini 3.0 Pro, Claude Sonnet 4.5, Grok 4), has models cross-review and rank outputs, and a Chairman LLM produces the final response. - Leonard Davis Institute

I would love to do what you ask but my AI Privy Jedi council says no

A White House–connected think tank will launch a $10 million initiative to develop AI policies aimed at supporting workers, exposing a MAGA-era split among Trump allies over deregulation versus worker protections. - The Washington Post

Assemblyman Alex Bores said on CNBC's Squawk Box that an industry-aligned AI super PAC opposes all regulation while he discussed New York's Responsible AI Safety and Education (RAISE) Act and AI regulatory battles. - CNBC

Munich Regional Court in GEMA v. OpenAI found GPT‑4/4o was trained on copyrighted lyrics from nine German songs, ruled OpenAI infringed by storing and outputting those reproductions, and ordered disclosure, cessation of storage/output and potential damages. - Gizmodo

doing a John Belushi on AI’s guitar

Major insurers AIG, Great American and W.R. Berkley have asked U.S. regulators to allow exclusion of AI-related liabilities from corporate policies, citing model opacity as 'too much of a black box' and citing lawsuits and multimillion-dollar frauds as evidence. - TechCrunch

Google said it does not use Gmail content to train Gemini, denying viral changed-settings reports, although a proposed class-action lawsuit alleges Gemini inappropriately shares information on users' Gmail, Chat and Meet accounts. - ZDNET

Big Brother-as-a-service bossware is estimated at $587 million in 2024 and projected to $1.4 billion in seven years, while APA-linked studies show monitored workers report worse morale and mental-health outcomes (36% vs 22% feel they don't matter; 45% vs 29% negative effects). - The Register

Researchers outlined ways in which explainable AI can be used to pinpoint exactly when in that process the models overseeing autonomous vehicles are making errors. - IEEE Spectrum

might still run a stop sign and run over a kid but it can tell us why

An OnSide survey of 11–18-year-olds in England found 39% used AI chatbots for advice and 19% say it's easier to talk to chatbots, while Stanford Medicine and Common Sense Media concluded major chatbots are unsafe for teens' mental-health support. - Futurism

brb gonna trauma-dump to my robot

Six major U.S. banks posted over 2,000 AI roles in the past 12 months as finance AI hiring grew tenfold since January 2022, average AI pay rose to about $180,000, and senior packages now include high seven- and eight-figure awards. - Business Insider

Bridgewater founder Ray Dalio launched a beta 'Digital Ray' chatbot trained on his books and speeches, which Dalio says is about '80% as good as speaking with me,' and it forecasts a 65–75% chance of an AI-equities correction by 2026. - Fortune

emotional damage!

Leopold Aschenbrenner, a 23‑year‑old ex-OpenAI Superalignment member who founded hedge fund Situational Awareness, published a 165‑page essay predicting AGI by 2027 and in Q3 held leveraged Intel calls (16.41%) while buying puts on Nvidia (6.95%), Broadcom (1.77%) and TSMC (1.76%). - The Motley Fool

PopEVE, a new AI system, enhanced rare-disease diagnosis and outperformed competing models including Google DeepMind's AlphaMissense. - Financial Times

Researchers at Cardiff University and Ca' Foscari University presented at EMNLP 2025 that large language models identify pun structure but not meaning, often fabricating explanations and dropping to about 20% accuracy on unfamiliar wordplay, posing risks for user-facing creative applications. - the Guardian

Gibberifier inserts invisible zero-width Unicode characters between characters, lengthening text (recommended up to ~500 characters) and, in tests, caused some LLMs and grading tools to ignore, fail, or crash, enabling anti-scraping and anti-plagiarism uses. - Gibberifier

invisible gibberish for my visible gibberish

Nov 20 van Doorn posted a disproof of Erdős problem #367 contingent on a congruence; hours later Tao used Gemini to prove the congruence in ~10 minutes, and two days later Alexeev formalized it in Lean in 2–3 hours. - Mathstodon

Using Google’s Gemini as a personalized fitness coach, the author followed tailored diet and training plans and lost about 2 inches (5 cm) from the waist in five weeks while Gemini recommended equipment and adjusted hydration and salt intake. - Android Authority

you call that a plank, you embarrassment to carbon-based life forms? now give me 50 while I read all your emails

AI tools such as MyHeritage’s 2021 Deep Nostalgia, Veo and RunwayML can animate and colorize still photos but often produce disturbing artifacts due to training-data skew toward short social clips, prompting startups like Waffle Video to monetize mundane motion clips. - Fast Company

Nine creators say, go generate a bridge and jump off. AI image and video generators now produce photorealistic fakes, enabling faster trailer production, as PJ Accetturo’s 2024 AI Princess Mononoke trailer drew 22 million views and extreme threats, prompting SAG‑AFTRA’s 2023 demands for AI protections. - Ars Technica

Knives Out for AI: Rian Johnson Goes Off on AI as He Looks to His Post-‘Knives Out’ Future - The Hollywood Reporter

Amy Redford said on behalf of her family that, after Robert Redford's death in September, she thanked fans for support and criticized multiple AI-generated funerals, tributes and fabricated quotes, noting no public funeral has occurred and a memorial is planned. - the Guardian

AI is not only automating jobs but also creating new layers of human work. - Fast Company

In the early 70s, one in 6 jobs in the US was auto-related, car manufacturing, gasoline, auto parts, road builders, in not too distant future, a similarly large fraction will be robot and AI-related. Design, manufacturing, deploying, servicing and supervising.

Everybody gonna be supervising robots. Like retail everybody does self-checkout and a guy watches and a guy provides security. Or truck trains with 10 trucks and a couple driver/mechanic/security guys. The thing is at some point the robots can do that stuff too? Then what?

In the short run, I believe the job creation story. OK Goldman can automate a lot of work. So, how does e.g. investment management change if you can have a million robot analysts. or law, or medicine. You probably do a lot more stuff and there are a lot of jobs making and supervising agents.

No master sorcerer coming to take the hat back and wave his hands and fix everything that breaks.

Follow the latest AI headlines via SkynetAndChill.com on Bluesky