AI Reading for Wednesday December 17

OpenAIOS? Altman is positioning ChatGPT to evolve into an operating system by adding app integrations (Peloton, Tripadvisor, Canva, Spotify), hiring Glen Coates as Head of App Platform, pursuing partnerships with Adobe and PayPal, and targeting a device for 2027. - Digital Trends

feels like Netscape wanting to run all your apps within the browser. But Google eventually made it fly with Web 2.0, Gmail, Gmaps, Sheets etc.

Trainium-spotting: Amazon is in talks to invest $10 billion or more in OpenAI, according to three people familiar with the discussions. Presumably Microsoft keeps OpenAI distro exclusivity. Vote of confidence in Trainium and not ideal for Nvidia although OpenAI already had deals for AMD and GCP and its own chips. - The Information

Better, faster fake images: OpenAI today launched GPT-Image-1.5 for ChatGPT users and the API, claiming up to 4× faster generation, ~20% lower cost, improved text rendering and small-face preservation, and an Images hub with preset styles for creative workflows. - OpenAI

OpenAI launched GPT Image 1.5, available to all users, which it says generates images up to four times faster and improves instruction-following and photo editing, and the company is positioning the model for enterprise use. - The Verge

Amanda Caswell tested OpenAI's updated ChatGPT Images against Google's Nano Banana across seven prompts and found Nano Banana won five categories, while ChatGPT won poster-text accuracy and multi-step character consistency. OpenAI responding but in reactive mode. - Tom's Guide

banana split decision: responding to Google but seem reactive and on back foot

OpenAI rolled back ChatGPT's router that gave Free and $5/month Go tiers access to complex models for complex questions, always defaulting to Instant, but allowing users to manually select deeper-thinking models. - WIRED

OpenAI Applications CEO Fidji Simo announced ChatGPT will gain Apple Music integration "soon," joining other 2025 app additions and likely enabling personalized playlist curation via Apple Music's public APIs while developers submit apps for review. - Apple Insider

OpenAI hired former UK chancellor George Osborne to lead the global rollout of its 'Stargate' initiative aimed at spreading a 'democratic' AI worldwide. - Financial Times

GPT 5.2 bombed in LMarena, falling behind 5.1. Artificial Analysis has it a tiny bit behind Gemini 3 Pro, a bit more expensive. - LMarena.ai

Roe vs. RAG: ChatGPT's abortion-related answers have fluctuated as it struggles to navigate the legal and political landscape, with Mashable finding in early December it flagged Texas access as "risky" amid HB7 confusion, while Plan C reported a 300% increase in ChatGPT referrals. - Mashable

Treasury CIO Sam Corcos approved procurement of up to 3,000 ChatGPT licenses, obligating about $1.5 million with over $500,000 already spent, under a Treasury user agreement that permits authorized mission use but restricts data and requires human review. - Fast Company

Google's 'Ironwood' TPU is over four times faster than its predecessor and, according to Morgan Stanley, TPU deployments could reach about 5 million units in 2027 and 7 million in 2028, with Broadcom reporting $21 billion in Ironwood orders. - Business Insider

Google working with Meta to make Pytorch fly on TPUs - Sherwood News

Google announces Gemini 3 Flash with Pro-level performance, rolling out now - 9to5Google

Say hello to my imaginary friend: Google cofounder Sergey Brin said he regularly talks to an unreleased Gemini Live model during his drives to ask technical questions, such as estimating data-center power needs and costs, and expects internal upgrades within weeks. - Business Insider

Move fast and break everything: Mark Zuckerberg's yearlong AI push at Meta drove colossal spending, fluctuating priorities and internal disorder that rattled company insiders and investors. - Financial Times

Meta mismanagement? - Beyond Search

Meta began rolling out Conversation Focus to Ray-Ban Meta and Oakley Meta HSTN glasses for Early Access users, using directional mics adjustable by swiping, and added a Meta AI–Spotify gaze-based music feature that plays music based on gaze. - The Verge

Meta released SAM Audio, a multimodal audio‑segmentation model on the Segment Anything Playground that isolates voices and instruments but cannot take audio prompts, struggles with similar sounds, and has no built‑in anti‑snooping protections. Meta simply said use must comply with law. - The Register

Oracle's $10 billion Michigan data center in limbo after Blue Owl balks - Yahoo

Project Sail, a proposed $17 billion, ~831-acre datacenter in Coweta County, prompted a moratorium and a draft ordinance for December 16, while Data Center Watch says $64 billion of AI datacenter projects have been blocked or delayed amid local opposition. - The Register

the big chill as data centers get cold shoulders from voters, financiers?

Outsourcing AI data centers to space poses technical, legal and logistical challenges that complicate feasibility and risk management. - Financial Times

That's heavy man: AI server racks now weigh about 5,000 lbs and draw up to 350 kW, versus 400–600 lbs 30 years ago, making retrofits often impractical and prompting plans for purpose-built datacenters, Uptime Institute CTO Chris Brown said. - The Verge

Databricks raised over $4 billion in a Series L at a $134 billion valuation and reported a Q3 annualized revenue run-rate of $4.8 billion (55% year‑over‑year), with investors including Insight, Fidelity, J.P. Morgan Asset Management, a16z, Blackstone and Temasek. - Fortune

Kirsty Craig was named a Tech Fellow at BlackRock and, as leader of a ~60-person portfolio-management research, data and AI team, helped build Asimov, an agentic AI virtual investment analyst deployed across BlackRock's $13.5 trillion platform. - Business Insider

Asimov's 2 laws of robot investing: 1) never lose money 2) never forget the 1st law

McKinsey plans workforce reductions of "a few thousand" roles—potentially up to about 10% in some non-client-facing teams—staggered over 18–24 months, and says its internal AI, Lilli, can cut search and synthesis time by up to 30%. - The Register

turns out AI can make slide decks telling McKinsey to lay off employees too

Microsoft reports 80% of leaders plan to integrate AI agents within 12–18 months; IDC predicts agentic AI adoption will triple in two years, and Frontier Firms already use AI across about seven functions. - Yahoo

Anthropic is testing Claude's new 'Tasks' mode, offering five structured workflows—Research, Analyze, Write, Build, Do More—and a persistent Project Manager sidebar to break tasks, track progress, and shift from chat LLMs to action-oriented agents. - Hacker Noon

Adobe updated Firefly with a prompt-based video editor that uses Runway's Aleph, added Black Forest Labs' FLUX.2 and Topaz Astra models, moved the editor from private beta to general availability, and is offering unlimited generations to subscribers until Jan 15. - TechCrunch

Adobe apparently reborn as an one-stop AI model integrator

Fei-Fei Li's World Labs released Marble in November and earlier RTFM, generating editable 3D world models from images, video or imagination for VFX, architecture, game development and robotics, with a partner reporting 40x faster ideation. - Financial Times

Mozilla CEO Anthony Enzor-DeMeo said Firefox will evolve into a modern AI browser over the next three years, with AI features, opt-outs, and promised user-control and privacy safeguards. - Windows Central

yay more AI no one asked for

DoorDash launched Zesty, an AI-powered social app pilot in the San Francisco Bay Area and New York that aggregates DoorDash, Google Maps and TikTok data and uses an AI chatbot plus social features to provide personalized restaurant recommendations. - TechCrunch

Waymo is seeking over $15 billion in funding at a nearly $100 billion valuation. - Bloomberg

A Waymo panicked and froze on a narrow bridge crowded with pedestrians during the Venice Canals Boat Parade in Los Angeles, stalling about 45 minutes, backing up traffic roughly a mile and causing viewers to miss its final boats. y no teleoperator? - New York Post

California’s DMV has ordered Tesla to fix misleading marketing, rename ‘Autopilot’ or face a sales suspension in the state. - Electrek

Europe can leverage withholding Dutch ASML lithography machines and enforcing EU data-protection rules to force US tech firms to rebuild AI pipelines, choke chip supply and slow US GDP growth, with AI accounting for about 92% of H1 growth. - the Guardian

Insane but so is sanctioning EU officials like terrorists and making it official US policy to end the EU

DeepSeek State: Chinese open-weight models like DeepSeek V3.2 and Alibaba Qwen have closed the performance gap with proprietary LLMs, prompting startups to adopt them to cut costs and protect data and spurring policy responses such as the White House AI Action Plan. - Techdirt

A 2023 concept called the 'jagged frontier' describes AI as having extreme strengths and weaknesses, with MIT-linked research finding about 95% of companies reporting zero ROI despite roughly $2 trillion in investment, complicating deployment and adoption. - Business Insider

An opinion piece cites Andrew Ross Sorkin's new book '1929' and 19th-century railroad booms to argue today's large AI investment surge resembles past speculative episodes and urges policymakers, investors and firms to balance deployment with risk management. - The Washington Post

First movers, exemplified by Roomba, can quickly lose market dominance as competitors overtake them. - Financial Times

Andreessen Horowitz Roadmap for Federal AI Legislation: Protect People, Empower Builders, Win the Future - Andreessen Horowitz

AI am the captain now: AI and a16z write the laws

Are companies using AI to rip you off? McDonald's says it monitors precise geolocation and browsing—and enable dynamic pricing; a 437-shopper Instacart test across four cities found nearly 75% of items varied up to 23%, costing some $1,200 yearly. - The New Republic

surge pricing for Big Macs

Daniel Kwan and more than 500 writers, actors and technologists launched the Creators Coalition on AI, publishing four core goals and gathering hundreds of signatories to set industry standards on transparency, consent, compensation, job protection, deepfake guardrails and creative safeguards. - The Hollywood Reporter

At Fortune Brainstorm AI, actor Joseph Gordon-Levitt criticized AI companies' self-regulation, urged legal guardrails, said generative AI's economic model favors tech firms over creators, and warned that AI 'synthetic intimacy' can harm children and drive dystopian outcomes. - Fortune

guardians of the galaxy, unite

A photorealistic AI-generated image falsely claimed the Bondi Beach Hanukkah attack was staged, was flagged by Google's Gemini SynthID, yet Grok, ChatGPT and Claude misclassified it as real; the real attack killed 15 and wounded dozens. - Gizmodo

Aura analyzed anonymized activity from about 3,000 children across roughly 90 chatbots and found 42% used AI for companionship, 37% of those engaged in violent roleplay (half involving sexual violence), with 11-year-olds’ rate at 44%. - Futurism

Asian bishops, led by Hong Kong's Cardinal Stephen Chow, said AI is 'a gift from God' not 'from the devil', urged sensible engagement with human oversight and safeguards, and rejected fully automated clergy tasks such as AI-generated homilies. - The Register

Terence Tao wrote on Mastodon that he doubts genuine artificial general intelligence is within reach of current AI tools, calling current systems a "clever magic trick" while others debate technical limits and possible advances within five to ten years. - Mathstodon

Big if true, AI beats the common cold:Researchers used AI to design nasal antivirals; KAIST's reengineered interferon‑lambda in nanoliposomes cut animal nasal influenza viral loads by over 85%, while WSU identified an amino‑acid fusion‑protein interaction preventing cell infection. - International Business Times

The Airpods have eyes: leaked iOS 26 prototype code indicates Apple may add cameras, ultra wide band to AirPods, with Visual Look Up, Contextual Reminders, a visual question-answering feature, and improved outdoor location understanding as soon as spring 2026. - Gizmodo

President Trump's 'Genesis Mission' collides with war on science. - Ars Technica

President Donald Trump filed a $10 billion defamation lawsuit against the BBC, alleging its Panorama documentary used AI to deepfake his January 6, 2021 speech to make it sound like he urged supporters to storm the U.S. Capitol, in a 33‑page complaint. - Futurism

you don’t need deepfakery to make some people sound stupid

Follow the latest AI headlines via SkynetAndChill.com on Bluesky