
live as a human. AI Lab @Google • ex-creator of agent systems @GoogleCloud • @Stanford CS/AI researcher
Mountain View, CA
Joined August 2016
- Tweets108
- Following208
- Followers1.5K
- Likes402
Pinned Tweet
I built a realtime interactive world. Just say it, and the world changes INSTANTLY.
Is it a new gaming experience? Open-world..
Support to render the real footage too as a filmmaker.
Links in reply 👇
Just say it, and the world changes INSTANTLY!
“Vibe” a world, make edits, and turn it a real footage.
> open source:
github.com/ryannli/spoken-wo…
> hero behind the scene: @GoogleDeepMind gemini hydrogen
Github & link to play: github.com/ryannli/spoken-wo…
Powered by our @GoogleDeepMind latest models, Gemini Hydrogen (fast world building), Gemini Transcribe (voice) and Gemini Omni (make it real)
2
8
830
Say it, and the world changes right in front of you, in your browser.
I open sourced “Spoken World”, a new open-world gaming experience, also can be used as a film set tools.
[models used]
> world building: Gemini Hydrogen
> voice input: Gemini Transcribe
> make it real: Gemini Omni
Github & link to play: github.com/ryannli/spoken-wo…
1
208
A super exciting opportunity for whoever obsessed with AI x creativity
I'm looking for a Student Researcher (BS/MS or PhD) to join me at @GoogleDeepMind Montreal!
We're working to get the best generative AI into the hands of the best creative people and pushing models past current frontiers for media & entertainment.
Links & how to apply 👇
1
3
599
I’m thinking a lot about human flourishing lately, seeing more research and engineer friends getting existential about AI.
I’m hosting a small-group workshop on this topic. If you feel resonated, welcome to connect, introduce & DM.
It’s not about helping humans to do more, but about helping humans become more —> meaning, creativity, connection, agency and fulfillment in the so called postagi era.
8
28
1,591
Welcome to carbon💎. I felt many shifts in my daily work already.
By the end of 2025, AI started crossing a threshold: working independently for 30+ minutes on well-scoped tasks. AI could finish the work we defined.
By the end of 2026, AI is starting to push forward work we haven’t fully defined. Open-ended autonomous work, figuring out what to do next alongside us.
The boundary of AI and human keeps shifting and it’s fascinating to find where it is.
Gemini 4 Carbon is our next frontier model.
It’s state-of-the-art across cybersecurity and long-horizon capabilities.
1
12
672
Building with genie 4 gives me tremendous joy. I visited many “places” that I never experienced in the real world before.
It could be super photorealistic, and the world is so responsive.
Past two weeks have been a blur. Four public talks, totally different topics, ridiculous deadlines. As an engineer, here’s how I quickly co-create my deck & storytelling with AI.
1/ Start messy and dump all contexts to AI. Talk for five minutes about what you might want to say. Dump in all the context: notes, docs, links, videos, audience, venue, time limit. No need to be coherent yet.
2/ Let AI interview you. Ask it to push on what’s unclear or contradictory. Stop when the questions stop being useful. Usually there are only 3–5 good ones.
3/ Get to the message before the slides. Ask for an outline, then ask: what are the three things people should remember a week later? If the answer feels off, fix the story first.
4/ Keep iterating on the structure, content, and length. Sometimes I ask AI to turn the five-minute version into a simple website, just so I can feel the story and vibe before making slides.
5/ Once the story works, build the deck with your own style / template to cut down on AI slop. Less is more. No text-packed slides.
6/ Bring in images, gifs and video, people love visuals! Gemini’s API key is super helpful to solve all.
7/ Talk through with AI what you want to say on each slide. Have AI shape it into speaker notes that fit the time.
8/ Generate an audio version (again, Gemini API key) and listen on your commute. Keep tuning the pacing and the story.
My main job: (1) deciding what I want to say / express, (2) judging what AI gives me, and (3) feeding it the taste, new references, and my “biases”.
1
15
671
I created a mini art work with Argon and Google’s earth data: Unknown similarities on earth
As an engineer who never learned art before, I feel blessed.
What I used?
- Data: Google Earth AI, Google Maps
- Music: DeepMind Lyria 3.5
- Crazy distributed data mining (over 500K locations, over 1 billion comparisons) & video editing: Gemini 4 Argon
I asked gpt-6 to dig through my “Photos” app on MacBook, blend fragments of different memories, and create something new artistically.
Here is what I get (and yes I can clearly recall where each fragment comes from!! 😯 such a special feeling)
Exciting night to demo Gemini Omni with @GoogleDeepMind team!!
Young talents kept throwing their great questions. My voice is shot.
And definitely saw a few jaws drop during the demos 😉





















