AI Reading for Wednesday March 11

Mystery CashApp transfer: Nvidia made a significant undisclosed investment in Mira Murati's Thinking Machines, which committed to buy at least one gigawatt of Nvidia hardware—about $50 billion of AI compute by Nvidia's estimate—and will begin deployment early next year. - SiliconANGLE

the $100b to OpenAI never quite materialized, these deals seem to have optionality

Microsoft announced a third-wave M365 Copilot focused on agentic AI with Copilot Cowork, launched Agent 365 (general availability May 1 at $15 per user/month), a $99-per-user M365 E7 bundle, and model choice including Anthropic's Claude. - TechRadar

Please don’t press ‘nuke’: Microsoft filed an amicus brief in favor of Anthropic getting a temporary restraining order against the Pentagon. - CNBC

Anthropic launched the Anthropic Institute consolidating three research teams and named co-founder Jack Clark head of public benefit, opened a Washington, DC office, and is suing the US government after being designated a DoD supply-chain risk. - The Verge

Google integrated Gemini into Workspace's Drive, Docs, Sheets and Slides, adding a chatbot-style prompt that pulls context from Gmail, Drive, Chat and the web to create and edit documents, auto-build spreadsheets, and offer suggestions Google says remain private until approved. - Ars Technica

gws “where is my email” - Google launches gws CLI letting autonomous agents manipulate Docs, Drive, Calendar, Gmail and Sheets, signaling shift toward agent-compatible interfaces over GUIs. - The Register

With NotebookLM you can ingest PDFs, web links, YouTube and Drive, performs Fast or Deep web research, create Audio Overviews with guest 'Join' participation, generate slides and flashcards, and integrate with Google Workspace and Gemini. - Lifehacker

Make me a TikTok: OpenAI plans Sora video AI launch in ChatGPT to drive user growth toward one billion - The Information

OpenAI's ChatGPT ad pilot attracts 100+ brands pledging $200,000+ budgets but advertisers report slow spend pacing eight weeks after February launch - Business Insider

QuitSkynet? Historian launches QuitGPT campaign urging ChatGPT boycott over Pentagon deals and lethal autonomous weapons concerns - the Guardian

Gracenote sues OpenAI for copyright infringement over media metadata (ironic if you remember how they appropriated all the CDDB IP) - Axios

Add this injunction to your shopping cart: Federal judge Maxine Chesney issued a preliminary injunction ordering Perplexity to stop Comet agents from placing orders on Amazon, destroy Amazon-derived data and refrain from accessing Amazon after finding "strong evidence" of unauthorized access, effective in seven days. - The Verge

Humans can no longer reliably distinguish AI-generated content from reality, performing near random chance - Financial Times

Nothing is real: AI-generated footage of fake Iran war explosions spreads on social media - CNN

YouTube expands AI likeness-detection pilot to journalists and politicians requiring ID verification - The Verge

Students deploy AI image-to-video tools to create defamatory 'slander pages' targeting teachers on TikTok and Instagram - WIRED

"Have fun shooting up the school!" Study finds chatbots assist with violent attack planning in 75% of test cases - the Guardian

Organizers say Washington State's SB 6346 public-comment record was flooded with about 37,824 extra opposition sign-ins, including over 15,000 identical repeated names, raising concerns that AI-generated mass fake comments distorted the process. - Techdirt

I’m just a bill, written by ChatGPT: Leaked memo says a top Senate administrator authorized aides to use ChatGPT, Google Gemini, and Microsoft Copilot for official Senate work, including preparing briefings. - The New York Times

China moves to ban OpenClaw on computers at banks and state agencies. - Bloomberg

Architect Nikhil Jathar built ERPClaw, a 45-module open-source MIT-licensed AI-native ERP with 385+ executable actions across 14 domains, aiming to undercut legacy SAP-level costs (about $50,000/year) for mid-market customers. - Hacker Noon

Open source ERPs and CRMs like Odoo and Corteza have been around for a while, I don't think a vibe-coded alternative will put SAP out of business. But it doesn't help SAP either. AI can supercharge nimble, low-cost competitors, and a few people who use 5% of it might code an in-house solution.

Oracle says AI code-generation tools boosted engineering productivity and helped Q3 revenue reach $17.2B (+22% YoY), with $4.9B in AI infrastructure (+84%), $553B RPO and $29B of contracts, enabling it to dodge the SaaSpocalypse. - The Register

Control-Z for rogue agents: Cohesity, ServiceNow and Datadog announced a recoverability suite to preserve immutable snapshots and enable point-in-time recovery of AI agents, models and data, and Gartner projects up to 40% of enterprise applications will include task-specific agents in 2026. - The Register

A Claude Code workflow with an opslane/verify pipeline—including preflight checks, an Opus planner, Sonnet/Playwright browser agents and an Opus judge—that produce code, run per-acceptance-criterion tests and return pass/fail/needs-human-review verdicts, yielding 40-50 PRs/week versus ~10. - Claude Code Camp

Agentic coding tools can perform routine refactors, lower prototyping cost to reduce technical debt and speed time-to-market, and improve through documented iterative prompts and operational practices that compound small quality gains. - Simon Willison’s Weblog

A three-factor model identifies generative-model use cases depending on relative encoding cost, relative verification cost, and artifact-versus-process dependence. Models suit low-prompt, low-verification tasks but poorly fit complex, safety-critical work and can cause economic harm if misused. - William J Bowman

or as the CEO says, can’t AI do this?

Mean girls, but for models: Startup Memvid posts $800 'Professional AI Bully' role to test memory failures in leading chatbots - Business Insider

AI summaries don’t pay the rent: Patreon CEO Jack Conte warned AI could create a "bloodbath for the world's creative people" unless tech companies pay creators, citing Anthropic’s $1.5 billion settlement and urging opt-out/payment rights and bipartisan training-data transparency. - Business Insider

Is Paramount’s AI-first merger a “force multiplier” or “'Flyboys' all over again”? - Vanity Fair