AI Reading for Wednesday December 3

Generative AI has moved past Gartner's Peak into the Trough of Disillusionment after rapid adoption, with ChatGPT over 800 million users, firms reporting AI-linked job cuts, productivity growth more than doubled versus the 2010s, and US employment ~5% below trend. - Financial Times

The chatbot wars intensify: Sam Altman declared a 'code red' as global rivals and new entrants, including Mistral's new models, narrowed OpenAI's early lead, intensifying competition that pressures OpenAI's market share and accelerates product rollouts and potential regulatory responses. - Financial Times

In 2022 Google declared a 'code red' over ChatGPT, now it's OpenAI's turn - SFGate

OpenAI Loses Key Discovery Battle as It Cedes Ground to Authors in AI Lawsuits. Must provide all communications discussing LibGen pirate book site. - The Hollywood Reporter

OpenAI launched a Shopping Research feature in ChatGPT, but Amazon blocked ChatGPT-related crawlers via robots.txt, causing ChatGPT to omit Amazon links and reflecting Amazon's effort to protect roughly $56 billion in advertising revenue and its Rufus shopping bot. - ZDNET

ChatGPT suffered an outage. - CNBC

OpenAI sources say they are developing a new LLM, codenamed Garlic, that outperforms Gemini 3 and Claude Opus 4.5 in coding and reasoning tasks, per internal evaluations. - The Information

Does it ward off vampires?

Investment in AI reached $252.3bn in 2024—down from a $360.7bn 2021 peak but 13 times the level a decade earlier—as expansion slows and Gartner warns firms must completely rethink business models to realise AI benefits. - Financial Times

Anthropic hired IPO law firm Wilson Sonsini as counsel as it prepares for an initial public offering that could be among the largest ever, signaling near-term filing and intensifying competition with OpenAI for public-market capital. - Financial Times

A profile of Dario Amodei and Anthropic, which is on track to hit a nearly $10B annualized run rate by the end of 2025, more than 10x what it generated in 2024 - - Fortune

can’t lose with that cardigan, 3 comma club here we come

Allowing AI to train itself. - the Guardian

Anthropic surveyed 132 engineers, conducted 53 interviews and analyzed about 200,000 Claude Code transcripts in August 2025, finding engineers use Claude for roughly 59–60% of work and report about a 50% productivity boost versus ~28% a year earlier. - Anthropic

The 9-person Anthropic team trying to stop AI from ruining the world. - The Verge

Millions of Coders Love Cursor. Can It Last? - The Wall Street Journal

If you haven't heard that dude talk, it's an experience, put him and Karpathy in a room and it's the bandwidth singularity.

Amazon Web Services launched Nova Forge, a $100,000-per-year service that lets customers access Nova models mid-training to inject proprietary data and optionally refine open-weight models, with the fee excluding training data, compute infrastructure and engineering assistance. - CNBC

Amazon: benchmarks, shmenchmarks. SVP Rohit Prasad said benchmarks do not reflect real-world utility, Amazon announced Nova Forge a service providing access to Nova model checkpoints at pre-, mid- and post-training to let enterprises customize models without massive cost - The Verge

AWS unveiled three developer 'frontier agents'—Kiro, a cross-session AI IDE that can run for days; a DevOps agent claiming 86% root-cause identification; and a Security agent for autonomous penetration testing—to automate triage, multi-repo code changes and security testing. - The Register

AWS in big Trainium push, following Google's TPU - DataCenter Dynamics

Amazon announced a new Trainium AI chip intended to compete with Nvidia and Google. - Bloomberg

Amazon Quietly Rolls Back Cursed AI Anime Dubs After Massive Backlash - Gizmodo

Salesforce says AI influenced about 20% of Cyber Week orders and its Agentforce drove 17% of partner online orders (~$13.5 billion); Adobe reports $11.8 billion on Black Friday and that AI-chat visitors were 38% more likely to buy. - Business Insider

Google is rolling out Gemini Agent, an experimental Gemini 3 Pro–powered AI assistant to Google AI Ultra subscribers in the US that uses live web browsing, multi-step orchestration and native Gmail/Calendar/Drive/Maps integration with human-in-the-loop confirmations. - AI Tools Club

Google is testing merging AI Overviews with AI Mode on mobile globally, letting users ask follow-up questions from Search results; AI Overviews reach about 2 billion monthly users and Gemini exceeds 650 million monthly users. - TechCrunch

Google plans to use what it knows about you to hyper-personalize AI and win the AI race - Gizmodo

Apple appointed Amar Subramanya as vice president of artificial intelligence, reporting to Craig Federighi and succeeding John Giannandrea (retiring spring 2026), to lead foundation models, machine‑learning research and AI safety. - Mashable India

Amar Subramanya, the 46-year-old Google and Microsoft veteran who will now steer Apple’s critical AI strategy. - Fortune

No pressure! Just make Siri do all that stuff we promised 2 years ago

Mistral AI released the Mistral 3 family—Ministral 3 (3B, 8B, 14B) and Mistral Large 3 (sparse MoE 41B active/675B total)—with Apache 2.0 weights,256K multimodel context, trained on NVIDIA Hopper/H200 GPUs and available on major cloud marketplaces. - Mistral.ai

within spitting distance of top Chinese models

Ashburn, Virginia, hosts about 152 data centers and is estimated to carry 70% of global internet traffic. Data-center construction rose to roughly $40 billion monthly in 2025, fueled by AI projects from Google, Amazon, Microsoft and OpenAI. - France 24

The US is considering using diesel generators to meet growing power demand from artificial intelligence operations. - Bloomberg

TSMC founder doubts Arizona fabs will work out in the long run, they are making subsidized wafers that need to be air-freighted to Taiwan for packaging into chips and circuit boards. - SemiAnalysis

A tech industry effort to block state AI laws stalled in Congress. - Bloomberg

ASPI reports China uses AI as a 'precision instrument' for censorship and surveillance, embeds it in law enforcement, exports these tools, and cites tests showing stronger censorship on political content. - The Register

AI Law and Border: Palantir's Immigration OS will find you and deport you - The Washington Post

A February survey of over 1,000 UK undergraduates found 92% use AI (up from 66% in 2024) and 88% use generative AI for coursework (up from 53% in 2024), prompting proposals for conversation-based, continuous AI-supported assessments to replace conventional exams. - Nature

Short-circuiting research with AI reduces learning. "when we focus on ideas in the company of others we encode the information in our brain more deeply and with greater understanding." - Harvard Business Review

How to prep your kids for an AI world - The New York Times

Randomized trials and pilot studies show AI ambient scribes improve clinician well-being and satisfaction but produce small documentation time savings, even as they attracted over $1 billion in investment this year and adoption by hundreds of health systems. - STAT

Other doctors say AI medical transcription saves them 1h per day - WHYY

AI-powered robo-advisors totaled about $1.4 trillion globally in 2024 and are projected to reach $3.2 trillion by 2033, offering lower fees (0.25–0.50% vs. 1–1.25% for human advisors) but raising algorithm transparency and data-privacy concerns. - Quartz

Waymo has been programming its self-driving cars to be "confidently assertive", which testing suggests has led them to bend traffic laws and show impatience toward pedestrians. - The Wall Street Journal

Hey! I'm walking here.

Waymo hits a dog in San Francisco. On Reddit, Waymo passenger says they are not sure a human driver would have avoided it, but would have reacted differently to a bump followed by panicked screaming from the back seat. - LA Times High School

’Fans have noticed issues with the newly rereleased Beatles Anthology documentary, including John Lennon's guitar appearing to have about 10 strings and Paul McCartney's face looking altered and "ghoulish".‘ - Bloomberg

AI Am The Walrus goo-goo g'joob

xAI engineer Parsa Tajik posted that he worked about 36 hours straight and shared a Tesla photo, drawing roughly 11,000 likes, praise from colleagues and a laughing-crying emoji from Elon Musk, while safety critics warned against the post. - Business Insider

If only they were robotic! Instead, chatbots have developed a distinctive grating voice. - The New York Times

Unitree robot playing basketball - X (formerly Twitter)

Schoolyard superman, he’s Jungle Jim

could we please not call it T800 tho?

Mark Manson is launching an AI Stoicisim coach intended to help people 'Subtly Not Give a F*ck'. - Fast Company

Follow the latest AI headlines via SkynetAndChill.com on Bluesky