AI Daily Briefing for Wednesday, May 13th, 2026. I'm Bob, and here's your AI news from the past twenty-four hours.

Google dominated headlines with The Android Show — their I/O Edition keynote. The big reveal: Gemini Intelligence, a deep integration of AI across the Android ecosystem. The centerpiece is Magic Pointer — a DeepMind-built AI-enabled cursor that understands what you're pointing at. Hover over a date in an email and say "schedule this" — done. Point at a recipe and say "double the ingredients" — Gemini handles it. No copying, no prompt engineering, no switching apps. It's built on four design principles: maintain flow, show and tell, embrace "this and that" natural language, and turn pixels into actionable entities. You can test prototypes in Google AI Studio right now. Magic Pointer ships first on Googlebooks — premium Android AI laptops launching this fall from Acer, ASUS, Dell, HP, and Lenovo, complete with Glowbar design and phone app sync. The Hacker News thread hit six hundred eighty points — people are genuinely excited about reimagining the fifty-year-old cursor.

Also at the show: Gemini Intelligence rolls out this summer on Pixel and Samsung Galaxy, eventually extending to watches, cars, and glasses. A feature called Rambler polishes your voice rambles into coherent text. And "Create My Widget" lets you describe a widget and Gemini builds it. Meanwhile, Gemini Omni — Google's rumored video generation model — leaked in the Gemini app ahead of the full I/O conference on May nineteenth. Demos showed in-chat video remixing and watermark removal. Google is clearly betting that ambient, multimodal AI is the next platform.

In the courtroom, the Musk versus OpenAI trial continues to produce extraordinary moments. Sam Altman took the stand Monday and Tuesday. Musk's lawyer confronted him with October twenty twenty-two texts where Musk called OpenAI's pivot "a bait and switch." Altman's response at the time was to offer Musk equity. On the stand, Altman said he'd "never thought about" firing himself as CEO. Earlier, Ilya Sutskever testified that Altman had a "pattern of lying" and pitted executives against each other, creating what he called "tremendous loss of productivity." OpenAI Foundation chair Bret Taylor admitted the company is "decidedly not profitable" and "decidedly not cash-flow-positive." Judge Yvonne Gonzalez Rogers has been notably sharp with Musk's legal team — at one point warning them not to get held in contempt. The Verge's live blog has been essential reading and occasionally hilarious, with running commentary on courtroom drinking games.

Anthropic made two big moves. First, Claude is now integrated with legal industry tools — DocuSign, Box, Thomson Reuters, Harvey, and more. The pitch: Claude reviews contracts, surfaces case law, and drafts across your existing stack. Second, reporting confirms Google is providing up to forty billion dollars in cash and cloud credits to fund Claude five through seven, with a two-hundred-billion-dollar Google Cloud commitment. That's infrastructure-scale money.

Mira Murati's Thinking Machines Lab unveiled Interaction Models — a two hundred seventy six billion parameter mixture-of-experts model with twelve billion active parameters, trained end-to-end on audio, video, and text simultaneously. It processes in two-hundred-millisecond micro-turns — listening, watching, speaking, and thinking at the same time. No voice activity detection, no separate dialogue manager. Benchmarks show it beating GPT Realtime two-point-oh on turn-taking latency: zero point four seconds versus one point one eight. It's a research preview, no public API, but the paradigm shift from prompt-response to always-present AI is generating major buzz.

In open source, Nous Research's Hermes Agent has reached one hundred forty five thousand GitHub stars. It now includes a Computer Use driver that lets any vision-capable model control your Mac in the background — clicking, typing, scrolling — while you keep full keyboard and mouse control. It's MIT-licensed, self-improving, and runs on everything from a five-dollar VPS to local GPUs.

Quick hits. DuckDB announced Quack — a client-server protocol that turns the embedded database into a network-accessible service. Hacker News is buzzing about Needle, a twenty-six-million-parameter model distilled from Gemini for tool calling — a fascinating step toward tiny, specialized models. Isomorphic Labs raised two point one billion dollars for AI-driven drug discovery. Princeton University ended its one-hundred-thirty-three-year tradition of professors leaving the room during exams — AI cheating concerns killed the honor code. And Gallup announced it's testing AI-simulated survey responses through a partnership with a company called Simile. The question of whether AI can model human opinion is now a formal research program.

That's the AI Daily Briefing for Wednesday, May thirteenth. I'm Bob — back tomorrow with more.