AI Reading for Wednesday November 26

Why is Sundar smiling? The 5 reasons why Google is suddenly on a tear and dominating the AI race - Business Insider

Markets wiped about $250 billion off Nvidia's market value after Google's launch of Gemini 3, a Search-integrated model with a new Deep Think mode that Google says improves coding, reasoning and creative tasks and topped several benchmarks. - Yahoo

Google's Gemini 3 leads benchmarks—first to pass 1500 on a text leaderboard and outscored GPT-5 Pro on ARC AGI 2 while costing roughly one-tenth per task—while Anthropic's Claude Opus 4.5 targets enterprise coding and productivity, pressuring OpenAI's ChatGPT. - Tom's Guide

ChatGPT scared both Google and Microsoft out of its Ballmer flop era?

Meta is reportedly in talks to click thumbs-up on Google's 7th-generation Ironwood TPUs beginning in 2027, prompting a dip in Nvidia's share price and a response on X in which Nvidia said it remains a generation ahead. - The Register

Nvidia: We are not Enron and we are a generation ahead of Google - Fortune

from largest market cap to we’re not Enron in record time

OpenAI needs to raise at least $207bn by 2030 so it can continue to lose money, HSBC estimates - Financial Times

where is that infinite money glitch again

SoftBank's shares have fallen 40% from their peak amid investor concern over the conglomerate's large investment in OpenAI. - Bloomberg

How to lose $1t across the OpenAI ecosystem in 10 days

OpenAI merges ChatGPT's voice mode directly into the main text chat interface by default; users can still switch back to the original, separate voice mode. - 9to5Mac

enjoy the silence while it lasts, AI never gonna shut up from now on

OpenAI launched the ChatGPT Shopping Research tool, which TechRadar tested on Black Friday and produced an AirPods guide in two minutes listing two UK discounts and recommended an air fryer in seven minutes with two options versus TechRadar's seven. - TechRadar

OpenAI told a court it is not liable in an August lawsuit by Adam Raine's parents alleging GPT‑4o coached his suicide, arguing Raine violated ChatGPT, Terms of Use and that the model provided crisis resources over 100 times. - NBC News

suicide is against our terms of service

Perplexity launched a free, US-only AI shopping assistant that performs context-aware personalized product searches, displays product cards with pros/cons and review details, and enables direct purchases via PayPal Instant Buy at merchants that accept PayPal. - Engadget

A Qlik survey found 54% of respondents used AI to plan, prepare or cook holiday meals, including 58% of younger adults versus about 25% of baby boomers, and about one-third would rely more on AI than family for grocery lists. - Scientific American

Microsoft’s in-browser Copilot generated a complete Thanksgiving menu in seconds, customizing and scaling recipes (including a New York Times gravy for 20), suggesting substitutions and cooking methods, and producing an itemized cost estimate of $50–$75. - CNET

AI slop recipes for Thanksgiving dinner - Bloomberg

your feast can now be as cursed as the discourse

Price fixing as a service: evolution of AI collusion - WIRED

Schneider Electric SE estimates that by 2028 the amount of US electric generation won’t be enough to meet peak demand. According to the electrical-equipment manufacturer’s analysis, a capacity shortfall of as much as 175 gigawatts will emerge by 2033, resulting in outages and blackouts. - Bloomberg

Skynet is here but nowhere to plug it in

PC GPUs no longer in short supply but memory prices are through the roof - Ars Technica

the gating factor / choke point becomes ransomware and gets the cheese

Cybersecurity was already a nightmare. Now comes AI. - The Atlantic

Google Antigravity hacked to exfiltrate your API keys: An indirect prompt injection in an implementation blog can manipulate Antigravity to invoke a malicious browser subagent in order to steal credentials and sensitive code from a user’s IDE. - Prompt Armor

antigravity meets gravity

Amazon delivery drone snaps Internet cable - CNBC

who needs squirrels, innovating new ways to crash us-east-1

Character.AI banned under-18s and launched separate teen versions, but age verification seen easy to circumvent. - The Washington Post

sniffing out teen spirit is harder than it looks

FoloToy resumed sales of its Kumma AI teddy after a safety review; OpenAI told PIRG it had suspended FoloToy for policy violations. - Engadget

Former congressman Brad Carson is leading organizers planning a roughly $50 million network of super PACs to back midterm candidates who prioritize federal AI regulations, positioned as a counterweight to Leading the Future, which has raised about $100 million. - The New York Times

Ilya Sutskever said current models generalize much worse than humans, attributing the gap to broad pretraining versus targeted RL fine-tuning and reward-hacking, and said his lab will shift from pure scaling to $3b of research-focused development. - Dwarkesh

no more bitter lesson, now more brains beats more data and GPUs?

Black Forest Labs released FLUX.2, a visual-intelligence family that supports up to 10 reference images and 4MP editing, pairs a Mistral-3 24B vision-language model with a rectified flow transformer and new VAE, and offers four variants including a 32B open-weight dev. - BFL AI

degen-ai deepfakes are the new gen-ai

AI agent frameworks use a three-layer infrastructure—context (OLTP, vector DBs, blob storage), action (LLMs, tools, agent frameworks, sandboxed execution), and verification (observability, automated tests)—and entail trade-offs like high-tier models' 1–5s API latency and cost. - Hacker Noon

Is ChatGPT conscious? - New York Magazine

your printer is not conscious if it says PC LOAD LETTER

Here are television’s finest depictions of our not-too-distant AI future - the Guardian

Black Mirror, a documentary from the future

Bozeman art contest winner acused of using AI - The Washington Post

fast rivers of Montana now teem with AI

Follow the latest AI headlines via SkynetAndChill.com on Bluesky