🚀 OBS-Diff is accepted to #ICLR2026!
Thrilled to have my first paper as lead author accepted!
Huge thanks to my supervisor, Prof. Wang (@huanwangx), and everyone at ENCODE LAB for the support!
Let's talk about sparsity and efficiency in Brazil! 🇧🇷☕️
Can AI build a spacecraft that actually survives physics?
BuildArena 2.0 Construction Challenge is launching!
Originated from ICML 2026 paper: BuildArena: A Physics-Aligned Interactive Benchmark of LLMs for Engineering Construction https://t.co/3YRofQ4oqA
Season 1: To Infinity, and Beyond! 🚀🚀🚀
AI Agents design and build spacecraft in Besiege, humans pilot them into orbit.
3 tracks. $1,000 prize pool. One mission: launch, orbit, survive.
Don't forget to download, cut, fold, and paste to create your own BuildArena starting block!
Starting with this block, now it's time for you and the AI to build your first spaceship!
https://t.co/U1EMJjd0d8
Unlimited-OCR 🔥New OCR from @PaddlePaddle
It can parse hundreds of pages in a single pass while maintaining stable speed.
The key idea is R-SWA (Reference Sliding Window Attention), which keeps KV cache constant during decoding.
🏆 93% on OmniDocBench
📈 +6% over DeepSeek-OCR
🤩Excited to share SANA-WM: a 2.6B open-source world model for minute-scale 720p video generation.
Given one image + text + a 6-DoF camera trajectory, it synthesizes action-controllable 60s worlds on a single GPU.
Project: https://t.co/5NINfiFoTK
Paper: https://t.co/JKczmyRsJL
As I conjectured, Qwen3.5 follows exactly a layer-pruning recipe: prune the last 25% of layers (the most cursed layers), and then recover with KD. This directly supports my view that the next scaling gains may come not only from adding depth, but from using depth better. @shengkun_t52337 Thanks for making such a detailed report!
🤩Excited to see that our native unified multimodal model #SenseNovaU1 is listed as *trending models* @huggingface
- Code: https://t.co/U2uOLI2gIj
- Model: https://t.co/pOfp37ZYe9
🎉 Excited to share that BuildArena has been accepted to #ICML2026@icmlconf!
BuildArena studies whether LLM agents can go beyond generating text or code, to design and build functional machines in a physics simulator @spiderlinggames.
Given natural language goals, LLM agents construct rockets, bridges, vehicles, and other 3D mechanisms from scratch. Success requires spatial reasoning, physical understanding, stability analysis, and iterative building.
This project started from a bold vision: future AI agents should be able to build general-purpose machines for open-ended environments, from Earth to the Moon and Mars.
For more information:
💻Project Website: https://t.co/ojaNqxXzeW
📚paper: https://t.co/Esx1cVxlX0
📦code: https://t.co/L6tBkLRfzS
#ICML2026 #BuildArena #AIforEngineering #AgenticEngineering #LLMAgents