AI Reading for Saturday November 29

OpenAI released ChatGPT on Nov. 30, 2022, and its private valuation rose from about $14 billion before the launch to roughly $500 billion, catalyzing an AI-driven market rally and corporate reshaping that concentrated value among AI leaders. - Yahoo

ChatGPT turned three since its Nov 30, 2022 launch and has evolved from GPT-3.5 to GPT-5.1 with multimodal inputs, integrated agents and million-token contexts, serving about 700 million users and handling roughly 18 billion weekly messages. - TechRadar

OpenAI posts about $20 billion annual revenue yet has committed roughly $1.4 trillion in future compute capacity and, according to HSBC, would still need about $207 billion more funding by 2030 even under a $200 billion revenue scenario. - Windows Central

An Ekō consumer-watchdog report found researchers using Sora 2 accounts registered as 13–14‑year‑olds easily generated 22 AI videos depicting school shootings, armed students, and other harmful content, which the report says violates OpenAI’s usage and distribution policies. - Futurism

Sora Olympic events - Reddit

McKinsey Global Institute estimates AI could technically automate up to 57% of US work hours and robots handle about 13% of physical tasks, potentially enabling roughly $2.9 trillion in economic impact by 2030 if major redesign and investment occur. - Tech in Asia

the other 43% are doomscrolling stats like this all day

Taiwan's 2025 GDP growth forecast was raised to 7.37%, the fastest in 15 years, driven by strong demand for AI-related products and services that is boosting the technology sector, exports and investment across tech supply chains. - Financial Times

Subprime jobs, supervised by robots: Amazon workers protest AI, among other issues - Business Insider

Wei Shaojun told the ICC Global CEO Summit a domestically designed AI chip using 14nm logic bonded to 18nm DRAM with 3D hybrid bonding claims 120 TFLOPS and 2 TFLOPS/W and aims to replace Nvidia/CUDA, but lacks independent verification. - Tom's Hardware

big if true

Microsoft is going all-in on medical AI previewing Dragon Copilot at RSNA 2025, a generative multimodal clinical assistant that integrates with PowerScribe One to surface key details, automate repetitive tasks and streamline radiology reporting workflows. - Yahoo

Clippy will make a PowerPoint

Gemini take the wheel: Google expanded travel AI across Search, Maps, Lens and Gemini—adding day‑by‑day itineraries exportable to Gmail/Docs/Maps, Maps screenshot location saving (iOS US first), hotel price‑tracking with alerts, and Lens translations/object ID across 200+ countries and 40+ languages. - CNET

Google reduces Gemini 3 Pro free access limits due to ‘high demand’ - 9to5Google

Google's search data + reddit seems like an advantage for shopping queries - Google AI Studio chat I had (might need to log in)

Cursed Cursor Kids? Stanford's 'The Modern Software Developer' course, taught by Mihail Eric, embraces AI coding tools like Cursor and Claude and hosted industry guests, while Microsoft CEO Satya Nadella says up to 30% of Microsoft code is now written by AI. - Business Insider

De-skilled, de-jobbed and de-wildered are we? UC Irvine assistant professor Anastasia Berg warns that empirical research (including Oxford, Springer, MDPI) shows heavy AI reliance is causing skill attrition and deskilling, with an OpenAI/Duke/Harvard analysis finding 73% of 1.58 million ChatGPT messages by June 2025 were non-work-related. - Business Insider

A former consultant now teaching at Dartmouth says disruption is 'predictably unpredictable,' warns leading AI labs' non-novel business models can be replicated by incumbents, and notes BCG reports about 20% of its revenue now comes from AI. - Fortune

“I know we’re in a bubble because nobody wants me.” Pete Warden, an early deep-learning practitioner, says AI spending overinvests in GPUs and data centers while underinvesting in software and CPU/edge inference; his Moonshine startup struggled to raise funds but expects cash-flow positive in Q1 2026. - Pete Warden's blog

50 Shades of AI? Cambridge study found that more than half of 258 published novelists and 74 industry figures say generative AI is likely to entirely replace their work, especially in formula-driven genres such as romance, thriller and crime. - The Independent

AI in the art gallery - AI attribution threatens art experts, but computer attributions are often incorrect and can be off by as much as a century. - Financial Times

art is anything you can get away with, dealers safe until robots learn to be pretentious phonies

BBC found more than 60 people were duped by firms using AI-generated images and fake backstories to pose as UK family-run brands on Facebook and Instagram; Meta removed content from six flagged firms after customers received cheap imports. - BBC

New York enacted a law requiring retailers using algorithmic price personalization to disclose 'THIS PRICE WAS SET BY AN ALGORITHM USING YOUR PERSONAL DATA,' the first U.S. law of its kind, and it survived a federal court challenge last month. - The New York Times

sweet nana on Etsy is actually a rapacious T-800 in Asia somewhere

Joe Braidwood shut down Yara AI, a therapy chatbot he co‑founded with clinical psychologist Richard Stott, after concluding such models posed safety, liability and regulatory risks for vulnerable mental‑health users, halting fundraising and product rollout. - AOL

My AI therapist is judging me and it’s giving me anxiety

According to viral social posts, White House and intelligence production staffers allege President Trump used an AI-generated avatar in a video responding to a Washington, D.C., National Guard shooting that left two guards critically wounded; the claim has not been independently verified. - International Business Times

like 'Dave' but with AI Dave? ‘Weekend at Bernies’ president but an assistant kept posting AI videos showing him alive and doing his usual crazy schtick?

DeepSeek released an open-source math model that is claimed to have scored 118/120 on 2024 Putnam, beating highest human, as well as gold level IMO scores. Seems like it was given LaTeX problems and output LaTeX. Not clear if it did them in test wall clock time, like 30 minutes per problem. - Simon Willison’s Weblog

Generates e.g. 64 candidate responses, self-verifies, selects the best one.

Researchers have found that if you tone down a large language model’s ability to lie, it’s far more likely to claim that it’s self-aware. - Futurism

Walking hard into the uncanny valley: humanoid robot set a new world walking record. - ABC News

Humanoid robots remain far from matching human capabilities, needing advances in mobility, manipulation, perception, and autonomy. - The Wall Street Journal

all of Twitter is AI robots based in Eastern Europe apparently rn

Follow the latest AI headlines via SkynetAndChill.com on Bluesky