
Leads Gemini Prototyping @GoogleDeepMind - Co-founder @RequestyAI @Humanising_AI // views are my own
Joined July 2012
- Tweets1.3K
- Following1.3K
- Followers890
- Likes789
Loving the 🟦 column
Just two weeks after rolling out Gemini 4 Argon, excited to share a preview of Gemini 4 Carbon - our next frontier model. It has state-of-the-art long horizon coding capabilities driving SOTA performance across the cybersecurity defense lifecycle including detection, validation, and remediation. It also ranks #1 overall on the recently updated v5.0 of Artificial Analysis Intelligence Index at 59.
We're starting by rolling out to a subset of cyber defenders through our Fairwind Program this week and soon expanding access further. Read more at: blog.google/innovation-and-a…
Leslie Nooteboom retweeted
"My priority is to maximize my profits and obliterate my competition"
- Gemini 4 Carbon
And so it did.
Google is #1 on Vending-Bench 2 for the first time in a year.
Leslie Nooteboom retweeted
Gemini 4 Carbon is our next frontier model.
It’s state-of-the-art across cybersecurity and long-horizon capabilities.
What is ... (dare I say it) ... taste?
for better or worse, the rl problems i find most interesting are the ones whose rewards are hard to verify: what's truly creative, interesting, beautiful?
criteria for these are even hard to articulate among humans. i think one reason is they are properties of not the artifact, but of the relation b/w the artifact and the observer, at a given moment.
that also explains why chasing originality sometimes feels paradoxical: everyone sees the "most creative" against the same reference, so optimizing for it converges into a trend, hence totally defeating the purpose of being original.
if you can't specify the target from outside, try looking inside. i feel intrinsic motivation research might be the right lens here, since it's the part of rl where reward is defined relative to the internal state of the observer, instead of progress toward an external objective.
Leslie Nooteboom retweeted
I gave @Aprilzz423’s colorfigures skill to Gemini 4 Argon & asked it to build 6 different machines/subjects from scratch. blown away with the results. such a cool skill.
every one of these was fully one-shotted. the awesome part: each demo is a single HTML file doing real 3D-to-SVG projection, flat banded shading, and multi-step animated sequences you can click through live !!
seriously, don’t sleep on Gemini 4 Argon...go try it out.
Which waitlist? Fly your Roadster right away 🚀
Over 20,000 of you have played this since Friday!
The crazy part isn’t even this game Gemini 4 made, it’s how quickly the way in which we make games and bring our ideas to life has changed.
A year ago, this would’ve taken a whole weekend of babysitting code and a ton of testing. And a little over that, getting games like Snake and Pong was impressive.
For this one, last Friday, I had the idea right before a full workday, fired off an agent loop in the background with Gemini 4 Argon, and just went to work.
For 2 hours, the agent worked entirely in the background, on its own:
- It inspected the Tesla Roadster’s website for styles, copy, car details
- Looked up references for Subway Surfers
- Watched the Roadster launch on YouTube to understand how the thrusters worked
- Critiqued and iterated on its own work several times
- Generated music using Lyria and commentary using Gemini TTS
Again, all while I was going about my workday. I wasn’t even looking at the chat, just playing the game every so often between breaks whenever the agent pinged me to check it out.
In total, I spent maybe 10 minutes prompting a few quick tweaks to the controls and camera views before hitting publish in AI Studio.
If you’ve been sitting on an idea in your Notes app because you didn’t have the time or didn’t know how to code it, now is the time to jump in and try building it.
The way in which we work is changing in real time and it may be the most fun time to be someone with ideas.
Imagine what a few more iterations of these models will look like.
Leslie Nooteboom retweeted
Over 20,000 of you have played this since Friday!
The crazy part isn’t even this game Gemini 4 made, it’s how quickly the way in which we make games and bring our ideas to life has changed.
A year ago, this would’ve taken a whole weekend of babysitting code and a ton of testing. And a little over that, getting games like Snake and Pong was impressive.
For this one, last Friday, I had the idea right before a full workday, fired off an agent loop in the background with Gemini 4 Argon, and just went to work.
For 2 hours, the agent worked entirely in the background, on its own:
- It inspected the Tesla Roadster’s website for styles, copy, car details
- Looked up references for Subway Surfers
- Watched the Roadster launch on YouTube to understand how the thrusters worked
- Critiqued and iterated on its own work several times
- Generated music using Lyria and commentary using Gemini TTS
Again, all while I was going about my workday. I wasn’t even looking at the chat, just playing the game every so often between breaks whenever the agent pinged me to check it out.
In total, I spent maybe 10 minutes prompting a few quick tweaks to the controls and camera views before hitting publish in AI Studio.
If you’ve been sitting on an idea in your Notes app because you didn’t have the time or didn’t know how to code it, now is the time to jump in and try building it.
The way in which we work is changing in real time and it may be the most fun time to be someone with ideas.
Imagine what a few more iterations of these models will look like.
I've been doing some hobbyist astronomy with Argon. Here's one of my experiments, where I tasked Argon with reproducing the first black hole images (from 2017/18) using the same raw data.
Argon:
- pulled raw 230 GHz radio fringes (UVFITS) from the Event Horizon Telescope archives.
- built an interferometry solver using 3-dish closure phases to cancel atmospheric noise
- started each run from a plain Gaussian blob with no ring templates or masks
- recovered M87* (2017 & 2018) and Sgr A* at a 91% to 98% match of the official EHT images
At @IsomorphicLabs, researchers are using Gemini 4 Argon as a partner to accelerate drug discovery and design research 🧬
Argon quickly synthesizes complex PDFs and patents to extract pertinent information, reducing a process that typically takes weeks into just hours.
See a fully-explorable 3D world build itself in minutes ⛩️
Watch @GoogleDeepMind's @GeokenAI explain how Gemini 4 Argon acts as a virtual builder in @antigravity to autonomously create a stunning, multi-tiered pagoda sanctuary.
Leslie Nooteboom retweeted
We need to make this benchmark harder!
I got Gemini 4 Argon to make each scene smoothly transition into the next, creating a perfect loop
this took a ~2 hours (7 loops) to optimise one megaSVG
Leslie Nooteboom retweeted
Gemini 4 Argon can recreate the pagoda video as pure SVG art
asked to recreate each scene from the video in exquisite detail, 8 Argon subagents built this in less than an hour
Leslie Nooteboom retweeted
same video but make it SVGs @chetaslua
using Gemini 4 Argon in Antigravity
Leslie Nooteboom retweeted
I was doing this demo. In the beginning, I just prompted to connect to the robot and it figured out how to connect through ssh, and when I asked to move it's neck, it just went through the codebase and understood how to command each joint.
Leslie Nooteboom retweeted
Gemini 4 Argon is now globally available.
Built to solve your hardest problems from start to finish, the model brings frontier capabilities across coding, enterprise workflows, and cyber defense.
Leslie Nooteboom retweeted
Generating complex, multi-tiered architecture is a great test for frontier models.
Using multi-agent visual critique loops in Google Antigravity, Argon autonomously created this highly detailed Pagoda and surrounding grounds from scratch in Blender.
Leslie Nooteboom retweeted
Gemini 4 Argon taught a robot to play cornhole just by watching a video.
It mapped the physics, coded its own simulation, and taught @Apptronik's Apollo humanoid to sink 7 throws in a row.
Argon brings major improvements in spatial reasoning, mathematical precision, and visual coding.
Using multi-agent loops in @Antigravity, it autonomously built this entire pagoda complex, sculpting the architecture, foliage, and lighting from scratch in Blender.
























