The TV box retired. Its remote sat in a drawer for six months. Now I hold its mic key, say "refactor this function," and my coding agent starts working.
300KB of Swift. No soldering. $0 in hardware. Open source on GitHub. https://t.co/6J57Rl8JwE
The TV box retired. Its remote sat in a drawer for six months. Now I hold its mic key, say "refactor this function," and my coding agent starts working.
300KB of Swift. No soldering. $0 in hardware. Open source on GitHub. https://t.co/6J57Rl8JwE
@ChenChaim42853 Exactly. That flat line told the truth when macOS was lying to my face. Great call on shipping the probe alongside the installer — adding a lightweight audio pipeline test tool to the repo so people can debug their own setup instantly.
Everything is open source — the BLE bridge, the mSBC decoder, the virtual mic driver, and notes on the four bugs that looked like witchcraft: https://t.co/Tc2QP8rg4n
Genie 4 的演示片,有三处把模型自己的思考过程打在了画面上。
热气球那条,右上角挂着 REASONING STEPS:
“Synthesizing the user's hot air balloon flight over Antarctica from the initial reference image and World State. … keeping the burner valve and vent line controls intact for future player interaction.”
注意最后半句:for future player interaction。它在生成这个世界的当下,就已经给"你后面要操作的东西"留了接口。视频生成器不用留接口。
丝绸那条更直白。用户输入"让风吹丝绸",系统回的是——"触发了已建立的世界规则:风速上升,绸子飘得更高、抖得更快"。改的不是这一帧,是往后所有的帧。
棋盘那条,左边标着 Gemini 4 Argon reasoning steps,把每一步落子写到"左手食指下压、向右下平移一格";右边标着 Genie 4 simulation。一个写动作,一个演。
所以它更像一个引擎的雏形,不是一段视频。
@sundarpichai The underrated unlock isn't the demo, it's evaluation: a world that stays consistent while you change one variable is the first place agents can be tested against something objectively gradeable. That's the part I hope gets documented.
@GoogleDeepMind The part I'm watching is determinism. In architecture, a client approves frame 4,200 and then asks for one wall moved — the whole sequence has to re-render identically except that wall. Is Genie 4 stable enough for that, or is every run a fresh sample?
This is Genie 4, our most realistic and interactive world model yet. 🌐
Explore dynamic environments in real time to see how the scene responds, sounds, and how the people in it move, react, and speak.