605 deaths before beating the Ender Dragon is very relatable.
Banger paper from Google DeepMind on long-horizon multi-agent coding.
(bookmark it)
They report that, to the best of their knowledge, this is the first end-to-end autonomous completion of Minecraft by AI agents.
They report that four Gemini 4 Argon agents beat Minecraft end-to-end. They defeated the Ender Dragon in 27 hours and 14 minutes with no human intervention.
Each agent controlled one player through the standard Mineflayer library. No one assigned roles, wrote game logic, or scouted a seed in advance. The agents planned, wrote, and debugged every program themselves, and coordinated through in-game chat and a shared memory.
Over the run, they wrote 2,783 programs and died 605 times. They split the work, shared what they learned, and recovered each other's items after deaths.
If you build agents for long software or enterprise tasks, this is a useful case study. The team had to keep one plan consistent across thousands of dependent steps for more than a day.
arxiv.org/abs/2610.22319
Chat with Paper: academy.dair.ai/papers/a-tea…
My favorite demo of Argon so far. This model has so many capabilities but one of them is is editing videos.
Everything from finding the right take, to constructing a narrative - this model can actually handles it all.
Gemini 4 Argon is insanely good at VIDEO EDITING and MOTION DESIGN 🔥
It understood my reference video and files really well because of its multimodal understanding, and looking at video frame by frame, code and image assets.
I had this unedited, unreleased video on my laptop, so decided to jazz it up and "re-imagine" it with Argon.
This video was entirely created inside Antigravity which is a great harness for creative work like this!
NOTE: this is not an official launch video, just my experiment.
anish retweeted
🤯 Wait, does this mean that @BenAffleck uses Veo / Omni / @nanobanana?!
anish retweeted
Ten years of Sad Affleck memes and it turns out he'd just been trying to install CUDA drivers the whole time.
Its very interesting to see how different models interpret "realistic"
GPT Image 2.5 on ChatGPT vs Nano Banana 2.1 on Gemini App
Prompt:
Create an ultra-realistic, high-end cinematic portrait of a beautiful young woman, framed in a tight head-and-shoulders composition. She faces the camera directly with a calm yet intense expression and a confident, captivating gaze.
Use dramatic low-key studio lighting with a strong directional light illuminating one side of her face while the opposite side gradually disappears into deep, atmospheric shadow. Keep the background completely black and seamless, creating a striking contrast around her silhouette.
Her eyes are deep brown with subtle geometric catchlights. Give her naturally textured, slightly damp-looking hair with individual strands visible and realistic volume. She wears a simple, elegant black crew-neck top with no visible logos or patterns.
Capture extremely realistic skin with visible pores, fine facial texture, subtle imperfections, delicate peach fuzz, natural lips, and detailed eyelashes. Avoid overly smooth or plastic-looking skin.
Photographed with a cinematic 35mm lens aesthetic, shallow depth of field, soft optical falloff, realistic tonal range, subtle analog film grain, and premium editorial photography quality. The final image should feel like an authentic photograph captured in a professional studio — sophisticated, raw, atmospheric, and completely photorealistic, with no obvious AI-generated appearance.
anish retweeted
nb 2.1 mogs image 2.5 for editing. huge unblocker for my video workflow
introducing @NanoBanana 2.1 🍌
our latest image generation model outperforms our previous models across the board:
- better visual design
- mask-based editing
- subject consistency
- more natural-looking images
try it out today in ai.studio
anish retweeted
👋 Hi folks!
I’m the new 🍌 PM.
Check out our latest image model! Nano Banana 2.1 outperforms our previous models across image gen and editing capabilities. And it makes great drawings of raccoons eating croissants!
Moved to the Bay to work on this, so hope you guys enjoy 😊
🤖 Made with AI
anish retweeted
Enjoy our latest model update, Nano Banana 2.1! It improves things tool use and instruction following (especially multi-turn), giving you more reliable and useful generations and edits.
anish retweeted
I'm in on this Nano Banana 2.1 cat meme. (bonus, that's my actual cat Miso.)
🤖 Made with AI
woah, the biggest upgrade in gemini 4 argon will be hallucination reduction
compared to 3.8 flash, google has basically crushed one of gemini’s biggest weaknesses
hallucinations now close to being eliminated
anish retweeted
Introducing Gemini 4 Argon – our new frontier model.
It’s built for complex workflows across coding, enterprise knowledge work, and cybersecurity defense – rolling out today to a set of trusted testers through our Fairwind Program.





















































