AI Reading for Thursday November 20

Nvidia reported Q3 diluted EPS $1.30 and $57.01 billion revenue, up 62% year‑over‑year, including $51.2 billion in datacenter sales, guided roughly $65 billion Q4 revenue, and saw shares rise about 5% after the results. - the Guardian

Nvidia reported Q3 EPS $1.30 and revenue $57.01 billion, with data-center revenue $51.2 billion, issued Q4 guidance of $65 billion ±2%, and shares rose over 5% after hours. - Yahoo

Chipmaker Nvidia reported Q3 revenue of $57 billion (up 22% Q/Q, 62% Y/Y), GAAP gross margin 73.4%, data-center revenue $51.2 billion, and CEO Jensen Huang said Blackwell sales are "off the charts" with cloud GPUs sold out. - The Register

Jensen has entered the chat - Reddit

party continues for another 3 months looks like

AI data-center expansion is sharply raising DRAM demand, prompting Samsung, SK Hynix and Micron to prioritize HBM and hike prices—Samsung up about 60% and some modules from ~$100 to >$250—straining device makers and risking shortages. - Yahoo

WaferLLM researchers at the University of Edinburgh and Microsoft Research developed three algorithms that let wafer-scale chips run LLMs about 100 times faster, cut latency roughly tenfold versus a 16-GPU cluster, and double energy efficiency. - The Conversation

Google is building custom Tensor Processing Unit ASICs in a restricted Silicon Valley lab and says it now spends more than $90 billion annually on AI build-out, while OpenAI projects about $1.4 trillion in commitments over eight years. - BBC

$1.4T in contracts but no audited financial statements. - Financial Times

An option is a contract but not a commitment. If buying a cheap option with a high notional value gets me e.g. 10% of AMD and they give me good terms b/c their stock rises on the announcement, I'll do that all day. Commitment is not full notional value. OpenAI and partners are not fools. But markets are foolish.

OpenAI updated the ChatGPT Atlas Mac browser with nine new features, including dynamic vertical tabs, an improved downloads interface, iCloud passkey support, Chrome extension import, an option to set Google as default search, and enhanced tab-management shortcuts. - MacRumors

GPT-5.1-Codex-Max offers better compaction, ability to handle bigger tasks, familiarity with Windows platforms - OpenAI Help Center

OpenAI introduced GPT-5.1-Codex-Max, an agentic coding model that coherently handles millions of tokens and runs over 24 hours, yields about 30% fewer thinking tokens, improves benchmarks (e.g., SWE-Lancer 66.3% to 79.9%) while rolling out in Codex with sandboxed defaults. - Simon Willison’s Weblog

400K context limit is smaller than competition

A draft White House executive order would direct the Justice Department to sue states that pass laws regulating artificial intelligence, following failed Republican congressional efforts to block state AI regulation. - The Washington Post

President Donald Trump on Truth Social attacked state AI regulation, claimed states embed DEI producing 'woke' models, and urged a single federal standard after a proposed 10-year state preemption provision was removed amid bipartisan backlash. - Raw Story

Josh Hawley, Ron DeSantis, Sarah Huckabee Sanders and other Republicans publicly opposed President Donald Trump's endorsement of a Silicon Valley–backed proposal to bar US states from imposing their own AI regulations. - Financial Times

Researchers found adversarial poetry achieved high attack-success rates against 25 proprietary and open LLMs—hand-crafted poems averaged 62% ASR, meta-prompt conversions of 1,200 MLCommons harmful prompts yielded ~43% ASR and up to 18x higher ASRs than prose baselines. - arXiv.org

practical use for poetry discovered, finally

French prosecutors expanded an inquiry into Elon Musk’s X to include Holocaust-denying comments generated by Grok that remained online for three days, drew over one million views, and prompted complaints from three ministers and human-rights groups. - the Guardian

AI-written Grokipedia cites Nazi forum Stormfront as a source 42 times, the conspiracy theory website Infowars as a source 34 times and the white nationalist website VDare 107 times, researchers found. - NBC News

Elon Musk said at the U.S.-Saudi Investment Forum that AI and robotics, exemplified by Tesla's Optimus robot, will make money "irrelevant" and work "optional," urging a universal high income while warning of significant disruption. - Business Insider

money will be irrelevant b/c Musk will have all of it and robots will patrol human work force with cattle prods.

Tesla reported three Robotaxi crashes in Austin in September, bringing the total to seven since launch after 250,000 miles, with NHTSA filings citing collisions with an animal, a cyclist and a backing vehicle, property damage and no serious injuries. - Electrek

Westinghouse partnered with Google Cloud to launch an AI platform integrating proprietary Hive, Bertha and WNEXUS tools to accelerate AP1000 reactor construction and operations, aiming for 10 units under construction by 2030 and reporting early pilot time and cost savings. - Google

DeepMind hired Aaron Saunders, former CTO of Boston Dynamics, to lead its robotics expansion and develop Gemini as an operating system for physical robots, aiming to integrate multimodal models for manipulation and logistics while addressing hardware and safety challenges. - WIRED

Google's Gemini app can answer 'Is this AI‑generated?' by detecting Google AI images using SynthID watermarks, will add video and audio detection soon, and plans to adopt C2PA metadata in Nano Banana Pro outputs and extend verification into Search. - The Verge

I gave Gemini 3 vs ChatGPT-5.1 9 challenging test — and Gemini won. - Tom's Guide

Using Gemini 3.0 Pro, demos turned a napkin sketch into a browser game - Geeky Gadgets

A study found that ChatGPT, Claude, Gemini and Meta AI often failed to recognize signs of mental health conditions in conversations and frequently provided general, non-specific advice rather than clinically tailored guidance. - The Wall Street Journal

Larry Page overtakes Jeff Bezos as 3rd richest as Google stock takes off - Bloomberg

SquareX researchers found a hidden MCP API (chrome.perplexity.mcp.addStdioServer) in the Comet AI Browser that lets embedded extensions, including Agentic, execute arbitrary local commands and are hidden from the extension dashboard, preventing user disabling. - Hacker Noon

Analysts and photographer William Vasta determined that a viral animated clip of a 2000 White House photograph of Donald Trump and Bill Clinton was AI-generated, with forensic team led by Manjeet Rege and Attestiv scoring it nearly 100% suspicious. - Poynter

China's transport and logistics sectors are rushing to deploy autonomous vehicles—pilot robotaxi services operate in about 20 cities, Baidu's Apollo Go reports 14 million rides, Cainiao 40 million driverless parcel deliveries—and forecasts warn over 7.5 million drivers could be displaced. - Financial Times

Uber Teams With Starship to Begin Offering Robot Deliveries in Europe - Bloomberg

Analysts warn a likely AI bubble—AI investment is about 17× pre-dotcom levels, NVIDIA valued near $4.6 trillion, and McKinsey found ~80% of companies report no significant earnings impact—risking startup failures and job losses while OpenAI, Google and NVIDIA likely survive. - Nature

El-Erian: A.I. Is a Bubble. Maybe That’s OK. - The New York Times

Companies halt stock buybacks to invest in AI - MarketWatch

Investors have surged into Oracle credit-default swaps, using them as a hedge against a potential AI-driven market crash. - Bloomberg

Oracle, initially favored by Wall Street for its AI prospects, experienced waning investor enthusiasm as reality tempered expectations. Somewhere, Safra Katz is laughing. - The Wall Street Journal

KPMG surveyed 753 Canadian business leaders and found only about 2% reported a return on generative AI investments, with ROI concentrated in firms with ≥$1B revenue; about 30% expect ROI within a year and 60% within 1–5 years. - Gizmodo

Lutnick family cashes in on AI - The New York Times

This scenario — a son making money on a project his father is boosting as a federal official — has come up repeatedly

U.S. national laboratories are accelerating efforts to integrate artificial intelligence with supercomputers, striking deals with tech companies such as Nvidia to speed development. - The New York Times

Major Wall Street firms ban or restrict applicant AI use while embracing AI internally, using detection tools and honesty agreements; TestGorilla flagged about 15% of over 5 million assessed candidates for potentially suspicious activity. - Yahoo

A history professor used a hidden "Trojan horse" prompt to detect AI use, finding 33 of 122 submissions flagged and 14 additional admissions—about 39% of papers partly AI-generated, with standard plagiarism detectors often failing to catch AI output. - HuffPost

LinkedIn is rolling out an AI mock interviewer that generates five questions, runs in an audio room, provides a readiness score and transcript feedback, will launch in English for LinkedIn Premium subscribers, and won't share content with hiring companies. - CNET

Organizations are prioritizing organizational intelligence in the AI era. - Fast Company

The EU's attempt to regulate AI highlighted tensions between establishing regulatory guardrails and attracting investment in Brussels. - Financial Times

Benedict Evans updated his May 2025 biannual presentation "AI eats the world," a long-form analysis of macro tech trends delivered to major corporate clients including Alphabet, Amazon, AT&T and Verizon. - Benedict Evans

Generative AI operates as a computational form of postmodernism, using simulacra, hyperreality and fragmented token-by-token outputs, and raising authorship, IP and deepfake trust concerns that prompt calls for provenance, watermarking and legal responses. - Hacker Noon

Reddit users reported ChatGPT converts text messages into images and sometimes claims to see non-existent uploads, community members recommend starting a new chat as a workaround, and the issue suggests a chat-state or generation bug requiring OpenAI fixes. - Reddit

me: animate this kitten

sora: For safety, we don't create videos from images that include people.

mission to animate Jensen failed I guess. Could try civitai which has a lot of cursed stuff.

Follow the latest AI headlines via SkynetAndChill.com on Bluesky