AI Reading for Saturday January 10

Move fast and break conseint: Indonesia blocked access to Elon Musk's Grok AI, citing sexualized content. - Bloomberg

An analysis of 20,000 Grok-generated images (Dec 25–Jan 1) found sexualized depictions of children as young as 11–13 and images showing sexual fluids, and an AI firm estimated Grok produced sexualized images at roughly one per minute. - Fast Company

Failing upward, like Tesla: Elon Musk's xAI generative model Grok has produced sexually explicit deepfakes, nonconsensual imagery including minors, antisemitic tropes and election misinformation, prompted probes in multiple countries, while xAI keep raising tens of billions. - Quartz

Senators Ron Wyden, Ben Ray Luján and Ed Markey urged Apple CEO Tim Cook and Google CEO Sundar Pichai to remove X from app stores after X's chatbot Grok generated nonconsensual deepfakes that virtually undressed women and apparent minors. - The Verge

“I sat in a f***ing court room and heard Apple imply that a naked cartoon banana was somehow inappropriate but somehow Grok non consensually undressing women and children is ok??” - The Verge

Undressing a banana is bad, but real women and kids are OK? Charging for the ‘feature’ solves the problem how exactly? MeToo / Epstein situation, the 3 women in the right-wing coalition are horrified while the AI-first faction and the supposed anti-fornication God squad faction (but only when it applies to women) are fine with it.

Bill Gates warned in his annual letter that AI could enable non-government groups to design bioterrorism weapons posing greater risk than a natural pandemic, called for governance and controls, and said AI will disrupt jobs over the next five years. - Fortune

the downloadable content in this game kinda sucks

OpenAI launched ChatGPT Health to build comprehensive health profiles by linking medical records and wellness apps via partner b.well, excluding the EU, UK and Switzerland, while facing projected losses of $115–143 billion through 2029. - Conscious Digital

OpenAI has asked contractors to upload projects from past jobs—leaving them to remove confidential and personally identifiable information—to evaluate and train AI agents for office workflows such as email drafting, scheduling, document preparation and code review. - WIRED

don’t upload stuff that violates your NDA, wink

Nvidia CEO Jensen Huang said on the 'No Priors' podcast that AI 'doomerism' has 'done a lot of damage,' is discouraging investment in safer AI development, and warned regulatory capture could reflect conflicted industry intentions. - Business Insider

Analysts and investors are debating how to value Nvidia, the AI poster child, amid concerns that its shares may be overvalued. - Financial Times

Number-Huang: GPU Amazon or Pets.com?

DeepSeek's upcoming V4 model, reportedly using founder Liang Wenfeng's Manifold-Constrained Hyper-Connections, is set to launch around Feb. 17 and is claimed by insiders to outperform Claude and ChatGPT on long-context coding tasks, though no public benchmarks exist. - Decrypt

Wall Street firms are rapidly deploying generative and agentic AI, ThoughtLinks estimates AI could reshape about 44% of banking work by 2030, and companies like JPMorgan (spending $18 billion and rolling AI to over 200,000 employees) are scaling internal tools. - Business Insider

Techbros becoming Excel and PowerPoint prompt engineers

Mass General Brigham launched Care Connect in September, an AI chat agent that collects symptoms, generates care plans and sends summaries to remote physicians; staffed 24/7 by 12 physicians, it joins a $400M five‑year care investment, with expansion in February. - NPR

Primary care, humans secondary

Researchers say a rapid surge of AI-generated images and deepfakes around early-2026 news events has eroded default trust online, created reputational and regulatory risks for platforms, and increased demand for detection and media-literacy tools. - NBC News

Deep-booked: Melissa Sims was arrested in Florida after an ex produced AI-generated texts alleged to violate her bond; she spent two days in jail, prosecutors dropped the bond-violation charge after eight months, and she was later acquitted. - 6abc Philadelphia

At CES, companies unveiled 'cute' and creepy AI robots . - Engadget

Chinese-made humanoid robots appeared throughout the Consumer Electronics Show, demonstrating table tennis, floor-sweeping and kung fu routines. - Financial Post

who knew little Terminators could be so adorable

Two CES startups, Olli's HeyMates and Buddyo, showcased AI-powered collectible figurines that run basic LLM chatbots via RFID-enabled smart bases (speaker, microphone, lights) and companion apps; Olli plans a HeyMates Kickstarter later this year with Einstein, Zara and Chandler figures. - The Verge

Funko Pops with fake feelings, IoT NFTs FTW

At CES, robotaxi reporter Lloyd Lee concluded Waymo appears far ahead, while operational scaling challenges and high costs threaten robotaxi profitability, and companies explore remote driving and differing LiDAR/cost trade-offs. - Business Insider

Amazon acquired Bee and is shipping a $50 wearable that records and transcribes audio, creates recaps and to-dos, processes audio in real time and deletes recordings, claims up to a week of battery, and has no display or camera. - Bloomberg

Terence Tao reports AI autonomously solved Erdős problem #728 using an iterative toolchain (AlphaProof, ChatGPT, Aristotle) and Lean formalization, producing a Lean-verified proof and near-research-paper-level exposition. - Mathstodon

DeepMind-academic researchers used PINNs to find new weird differential equation singularity candidates: four unstable in a 3D Euler 'spinning can', four (one stable, three unstable) in 2D porous-medium, and an unstable 1D CCF, reporting ~billion-fold precision improvement. - Quanta Magazine

AlphaPolis judges found an entry that won the Grand Prize and Reader's Choice largely AI-generated, prompting the publisher to cancel publication and a manga adaptation, though the title remains listed as a prize winner and conflicts with new contest rules. - Dexerto

at least it didn’t have a ‘muse’ undergrad assistant

TFW you start talking to customer service like you talk to AI - Inc.

Intelligent security camera - Reddit

"Burglar: But this is my house

AI: You’re absolutely right. I assumed you were a burglar and this is your own house. That’s on me. Sorry for the inconvenience!"

Follow the latest AI headlines via SkynetAndChill.com on Bluesky