AI Reading for Friday December 19

CE-No: OpenAI CEO Sam Altman did the Alex Kantrowitz podcast, said the company launched GPT-5.2, saw ChatGPT weekly active users double from about 400 million to roughly 800 million, committed about $1.4 trillion to long-term compute and projects about $20 billion in revenue this year. Being CEO of $1T sounds like the worst tho. - Big Technology

OpenAI is aiming to raise as much as $100 billion at $830 billion valuation by end of Q1, as it seeks to pay for ambitious growth plans in a market that has cooled recently on the artificial-intelligence boom. - The Wall Street Journal

OpenAI, valued at $500 billion in an October sale and seeking $750 billion, is preparing for an IPO possibly as early as H2 2026, while CEO Sam Altman said he is "0% excited" to be a public-company CEO. - Business Insider

Sources say tensions between OpenAI's research and ChatGPT product groups triggered a 'code red' and OpenAI is set to beat its $13 billion 2025 revenue goal. - The Information

OpenAI president Greg Brockman said the company is constrained by compute after reallocating capacity from research to deployment following a viral March image-generation launch, prompting accelerated data-center investment amid Altman's 'code red' push. - Gizmodo

OpenAI ships Codex 5.2 - OpenAI

We introduce evaluations for chain-of-thought monitorability and study how it scales with test-time compute, reinforcement learning, and pretraining. - OpenAI

AI psycho-analyzing its own inner dialogue now

OpenAI added a pinned-chats feature to ChatGPT that lets users fix conversations at the top of chat history, speeding retrieval of important threads for ongoing projects and reducing time lost to auto-labeled chats. - 9to5Mac

OpenAI’s new motto: losing money on every token but we gonna make it up on volume - X (formerly Twitter)

Anthropic launched Agent Skills as an open standard with a spec and SDK, announced Microsoft adoption in VS Code and GitHub plus partners including Atlassian and Figma, and said Claude aided 60% of engineers' work with a 50% productivity boost. - VentureBeat

Blog post - Anthropic

Anthropic: Safety measures we’ve taken to date, and how well Claude currently performs on a range of evaluations. We focus on two areas: how Claude handles conversations about suicide and self-harm, and how we’ve reduced “sycophancy”—the tendency of some AI models to tell users what they want to hear, like, you should definitely call your ex. - Anthropic

OpenAI, Anthropic gonna guess your age - The Verge

get carded by Clippy

Gap between Nvidia and Huawei may be widening, analysts say U.S. chips are ~5× more powerful today and projected ~17× by 2H2027, and Nvidia expects about 4–5 million H200s in 2025. - Council on Foreign Relations

China is hot-rodding older ASML lithography machines to produce more advanced AI chips, boosting domestic chip output, reducing reliance on Western suppliers, and expanding capacity for cloud providers, datacenters and AI model training. - Financial Times

Nvidia internal memo says it discussed AI‑infrastructure alternatives, including in‑house "AI factories" and neoclouds, with Capital One, which remains committed to AWS but is exploring alternatives to control rising AI-related cloud costs. - Business Insider

what’s in your wallet and is it enough to pay the AWS bill?

UBS forecasts global AI capex of $423 billion in 2025, $571 billion in 2026 and $1.3 trillion by 2030, while Alphabet holds nearly $100 billion cash enabling internal funding and Oracle faces heavy leverage after a ~50% stock decline. - Yahoo

Texas's AI law, taking effect Jan. 1 , drew opposition from 16 Texas senators after Trump's order threatened to cut broadband funds over state AI laws. - The Texas Tribune

President Trump issued an executive order launching the Genesis Mission, led by the Department of Energy and OSTP, to centralize federal scientific data and supercomputing resources into the American Science and Security Platform for AI-driven research. - Hacker Noon

Google DeepMind released Gemma Scope 2, an open-source interpretability suite for the Gemma 3 family (270M–27B parameters) that required ~110 PB of stored data and over one trillion parameters trained, intended to accelerate AI safety research and auditing. - Google DeepMind

an observability microscope for the little AI brain-worms

Google Labs launched CC, an invite-only Gemini-powered tool that scans users' Gmail, Calendar and Drive to generate daily "Your Day Ahead" briefings; the waitlist is open to personal Google accounts in the U.S. and Canada (18+), excluding enterprises. - Business Insider

maybe it can doomscroll for me

Hannah Fry and Demis Hassabis, The Future of Intelligence - YouTube

2 galaxy brains walk into a bar, are totally brilliant and adorbs

Beware of slow, confused, ‘Mad Max’ Waymos - Reddit

Tesla reintroduced 'Mad Max mode' in its FSD software, which drives aggressively (video showed 82 in a 65 zone) and prompted an NHTSA safety probe, while Waymo has also tuned vehicles to be more assertive amid legal and regulatory scrutiny. - Fast Company

hey the speed limit is just a suggestion

LG will let TV owners delete clingy Microsoft Copilot from their televisions following customer outcry. - Bloomberg

Firefox will offer an option to completely disable all AI features, dubbed the "AI kill switch", and make AI features opt-in, including a local ~20MB image-captioning model for alt-text, on-device translation, automatic captioning, text-to-speech and a sidebar chatbot. - Mastodon

i just wanna browse, MF, I don’t remember asking AI a goddamn thing

Anti-AI browser Waterfox announced it will not include large language models, positioning its Firefox fork as a privacy- and customization-focused alternative that tracks ESR releases, disables Mozilla telemetry, preserves XUL add-ons, and aims to attract users opposed to Mozilla's AI direction. - The Register

London Business School professor Ekaterina Abramova warns rapid AI disruption could, within five to ten years, cause layoffs that outpace new job creation as single AI models may displace thousands of cognitive roles, risking social unrest without mitigation. - Business Insider

Multi-agent systems are predicted to scale across enterprises in 2026, replacing single-model copilots with coordinated agent swarms that automate end-to-end workflows—e.g., Grove AI in clinical operations—initially adopted in rules-heavy functions and requiring rigorous guardrails. - Hacker Noon

get replaced by a swarm

YouTube removed two channels, Screen Culture and KH Studio, that had amassed more than 2 million subscribers by posting AI-generated fake movie trailers, after demonetizing them and requiring disclaimers in early 2025 and later restoring monetization. - Ars Technica

It's not you, it’s GPT, the 'it's not X, it's Y' structural tic as an AI giveaway. Sentence structure and rhythm, rather than specific words or punctuation, are stronger indicators of AI-generated text, driving demand for human post-editing and detection or repair tools. - Inc.

The article defines 'borrowed certainty' as users' confidence in AI answers received without verification, warns this reduces understanding and risks degraded expertise, judgment calibration, and accountability, and recommends norms, training, and design safeguards. - Psychology Today

AI-developed oral pill zasocitinib met endpoints in two late-stage trials, with over half of plaque psoriasis patients achieving clear or almost clear skin after 16 weeks and the drug outperforming placebo and Amgen’s Otezla. - Yahoo

The Cybertruck, sycophantic AI, and humanoid robots all made this year’s list of the biggest technology failures. - MIT Technology Review

Give me slop, Beautiful slop, The bubble is expanding And it's never gonna pop - Reddit

Follow the latest AI headlines via SkynetAndChill.com on Bluesky