AI Daily Dev — October 30, 2026

DeepMind Ships Genie 4, Its First Real-Time World Model With Native Audio

Oct 27: experimental research preview that generates interactive 3D scenes you can steer in real time with text prompts or keyboard controls.
First Genie model to produce audio natively — ambient sound, music, and character dialogue render alongside the scene instead of being post-added.
Scene continuity roughly doubles over Genie 3 to about two minutes; a smaller Gemini model observes the frame-by-frame output and guides the simulation.
Rolling out as a trusted-tester preview on Google AI Ultra's Project Genie, with Gemini API and AI Studio EAP next; all output carries SynthID watermarks.
DeepMind flags object persistence, long-horizon coherence, and complex physics as known weak spots — the lab is framing it as agent/robot training infra, not a consumer product.

models deepmind.google

+1,930 stars today lifts the HKU Data Intelligence Lab's open tutoring app past 50K stars, Apache-2.0, with a Next.js workspace and a CLI.
Dual-loop multi-agent architecture pairs a TutorBot with RAG over your own study materials; persistent memory tracks what you've learned across sessions.
Supports 30+ model providers (OpenAI, Anthropic, DeepSeek, Gemini, Ollama) with Docker self-host; UI runs on port 3782, no vendor lock-in.
Growth curve stayed sharp after hitting 10K in 39 days — the first non-coding agentic skill app to top the daily trending board this month.

open-source github.com

Oct 27: standalone Mail and Calendar plugins for Apple accounts land on Free, Plus, Pro, Business, and Enterprise — web, desktop, and iOS.
Mail can search inbox, draft, and send; Calendar can read, schedule, and update events; both can look up iCloud Contacts (no standalone Contacts/Reminders plugin yet).
Setup needs an app-specific password per plugin; Mail is limited to @icloud.com / @me.com / @mac.com — other accounts in iOS Mail (Gmail, Outlook) stay invisible.
Lands ~14 months after the Gmail / Google Calendar connectors and feeds into OpenAI's Dots proactive agent for the same accounts.

tools openai.com

TestingCatalog reports Carbon is now in a public LMArena rotation with Project Carbon EAP for select Gemini API partners starting this week.
Internal Googlers testing the checkpoint describe coding quality in the Claude Opus 5.5 ballpark on long agentic tasks — impressions, not published benchmarks.
Logan Kilpatrick's prior remark that Argon stays the near-term GA model lines up with this: Carbon is still a research checkpoint, naming (Carbon vs Argon-successor) is unresolved.
No pricing, context-window, or model-card page yet; Google has declined to comment on the leaks.

models testingcatalog.com

+1,034 stars today puts Jesse Vincent's agentic skills framework over 300K total — one of the fastest-growing MIT-licensed repos of the year.
Composable skills + a strict TDD-and-review methodology; subagents re-check each task before shipping, with /plugin install superpowers@superpowers-marketplace as the two-line Claude Code setup.
Now installs cleanly into Codex CLI, Cursor, Gemini CLI, GitHub Copilot CLI, and Antigravity — the same marketplace install, not forks.
Main critique in community threads this week: the mandatory process adds friction on trivial one-liners, where the plan+review loop is more overhead than the fix is worth.

open-source github.com

Research preview from Block (Jack Dorsey) that takes audio in and emits audio out — no STT, no LLM, no TTS layer between the two.
Full-duplex: can listen and speak at once, interrupts, laughs, and responds to emotional cues; demos cycle on block.xyz/dessa with a public waitlist.
Trained from scratch on human conversation on a single GB200 NVL72 rack; Block claims inference runs faster than real time on a MacBook Pro M4 Max (unverified).
Community reaction split — the demos read as uncannily natural to some reviewers and 'scary' to others; early-access is researcher-only, no weights or API yet.

research block.xyz