AI Reading for Friday December 12

OpenAI released a new model intended to surpass Google's recent product, as it faces a strong competitive challenge in AI. - The New York Times

OpenAI launches GPT-5.2, says it's better that Gemini on many benchmarks. - OpenAI Help Center

Also 40% more expensive per token in the API, although OpenAI claims it will output better results token-for-token, offsetting cost. 3rd-party benchmarks awaited.

Open source thought police: OpenAI launched GPT-OSS Safeguard, a reasoning-based mediation and moderation model for its open-weight GPT-OSS lineup, published on Hugging Face in 20B and 120B sizes that evaluates prompts and applies developer policies at inference while exposing its chain-of-thought. - Hacker Noon

Denise Dresser, who led Slack integration after Salesforce's $27.7 billion acquisition and served 14 years at Salesforce, is leaving to become OpenAI's first chief of revenue, overseeing global revenue strategy as Rob Seaman becomes interim Slack CEO. - Fortune

OpenAI's Fidji Simo says ChatGPT's “adult mode” is expected to launch in Q1 2026, says OpenAI wants to get better at the guess-your-age game before launching the feature - The Verge

Beauty and the bots: Disney to Invest $1 Billion in OpenAI, Allow it to Use Disney Characters in Major Deal That Boosts Sora In Hollywood - The Hollywood Reporter

make your own Disney movie, in ‘adult mode’, what could go wrong 

Bob Iger says the OpenAI deal “does not in any way represent a threat to the creators at all” because it doesn't include name, likeness, or character voices - The Hollywood Reporter

Disney will invest $1 billion in OpenAI under a three-year licensing deal allowing OpenAI's Sora to generate up to 20-second videos using about 200 Disney-owned characters from Marvel, Pixar, Star Wars and animated films while excluding talent likenesses and voices. - the Guardian

Disney and OpenAI agreed a three-year licensing deal letting Sora and ChatGPT Images generate user-prompted videos and images using more than 200 Disney/Marvel/Pixar/Star Wars characters (excluding talent likenesses/voices), with creations starting in early 2026 and Disney investing $1 billion. - OpenAI Help Center

Deep orangered-yellow-blue-green: Google relaunches Gemini Deep Research to be more powerful and accessible via API - Google

Scrapes of Wrath: EU competition regulators opened an antitrust probe into Google for using web publishers' content and YouTube uploads to train generative AI and power features, investigating unpaid use and privileged access, with fines up to 10% of global annual revenue. - Fortune

One nation under GPT: Trump signed an executive order purporting to create a single federal regulatory framework for artificial intelligence, that seeks to preempt state AI laws, gives the attorney general authority to sue states and directs federal agencies to withhold funds from noncompliant states. - The New York Times

In the beginning, God created compute. The Department of Energy awarded over $320 million under President Trump's "Genesis Mission" to build an integrated American Science and Security Platform, funding AmSC, a Transformational AI Models Consortium, robotics/autonomous lab projects, and 37 foundational AI awards. - The Register

Pre-emption may threaten national security by deregulating AI and social media and exposing them to foreign adversary attacks. - The Atlantic

bug or feature?

Stanford researchers developed AI hacking bot Artemis and say it surpassed nine out of 10 penetration testers by rapidly finding vulnerabilities in the university's network - The Wall Street Journal

China’s censorship and surveillance were already intense. AI is turbocharging those systems - CNN

China expanded its government-approved AI hardware supplier list naming domestic firms including Cambricon and Huawei, excluding foreign suppliers such as Nvidia and AMD, to govern procurement across ministries, state-owned companies and public institutions. - Tom's Hardware

Broadcom CEO Hock Tan reveals that Anthropic placed a $10B order for Google's Ironwood TPU racks in Q3 and says it placed an additional $11B order in Q4 - CNBC

$10b here, $11b there, and pretty soon you’re talking about real money.

Cooling the CUDA’s nest: High-density GPU racks drawing 40–60 kW and requiring liquid cooling above about 35 kW are overwhelming traditional data-center power and cooling designs (sized for 10–15 kW), forcing direct-to-chip cooling, lossless fabrics and about 20–25% oversized cooling. - Hacker Noon

Runway announced GWM-1, a family of three autoregressive "world models" built on its Gen-4.5 text-to-video foundation and post-trained with domain data, claiming minutes-long coherence, real-time environment exploration, and synthetic training data for robotics and simulation pipelines. - Ars Technica

Scaling laws of physics: Tim Dettmers of the Allen Institute and CMU says AGI is unrealistic because GPUs big gains in performance-per-cost peaked around 2018, scaling may have one to two years left, and rack-level stitching only extends gains until about 2026–2027. - The Register

SaaS with extra steps: Salesforce is adopting seat-based AI licensing through its Agentic Enterprise License Agreement, combining seat-count pricing with reusable credits and embedded AI-unit caps, a shift Marc Benioff says multiplies monetization three- to fourfold while analysts warn of buyer lock-in. - The Register

xAI, owned by Elon Musk, will deploy its Grok chatbot to more than 5,000 Salvadoran public schools serving over one million students over the next two years to provide AI-powered curricula, despite Grok's record of producing antisemitic and conspiratorial content. - the Guardian

like Ernest goes to jail but Grok goes to CECOT

Surge pricing but for groceries? Instacart is charging prices as much as 23% higher to different customers in a dangerous AI experiment, report says. “American grocery shoppers aren’t guinea pigs, and they should be able to expect a fair price when they’re shopping.” - LA Times High School

Cursor, the wildly popular AI coding startup, is launching a new feature that lets people design the look and feel of web applications via chat. - WIRED

Time Magazine named the "Architects of AI" its 2025 Person of the Year, featuring tech leaders including Nvidia's Jensen Huang, OpenAI's Sam Altman and Meta's Mark Zuckerberg, and warned of rapid AI deployment and associated risks. - BBC

enshittogenic policy environment of the enshittocene era, the real treasure was the artificial friends you made along the way

Follow the latest AI headlines via SkynetAndChill.com on Bluesky