Xingyi Yang

@yxy2168

Assistant Professor @ PolyU DSAI PhD @ NUS. Visiting PhD at @ Oxford. ML & CV

Hong Kong

adamdad.github.io

Joined September 2018

563 Following

1,719 Followers

Replies Reposts Media

Pinned
Xingyi Yang
@yxy2168
Sep 2
We share H3-World 🌍 The first to turn MiniMax-H3 itself into world model. No new action module. We directly convert H3’s pretrained language understanding into world control. Only 8K samples + 0.199% Trainable Params. 📄 huggingface.co/papers/2609.01… 💻 danzer1xxxxchan.github.io/H3-World/
00:00
41
Xingyi Yang
@yxy2168
Oct 1
Too many people claim that coding LLMs can handle 3D, but how exactly? We need more benchmarks on this.
Refresh
@refresh_dev
Sep 30
Introducing Blender Bench v1 (BBv1): a benchmark for frontier LLMs on 20 realistic 3D production workflows including animation, modeling, rigging, cloth & surfacing. GPT-6 Astra leads overall; Claude Opus 5.5 wins modeling & cloth simulation.
00:00
Xingyi Yang
@yxy2168
Sep 28
Coding LLMs are becoming a really powerful hammer. I’m curious how many 3D vision problems can now become nails once we give them code, tools, rendering, and visual feedback. Kind of love this trend.
00:00
Congrong Xu
@CongrongX
Sep 27
00:29
LLMs can already build a Blender scene from a room video. So is 3D vision doomed? We tried the opposite: hand the vision models to GPT-6 as tools (Pi3X for geometry, SAM3, GVHMR/Kimodo for people). It turns a room video into a better interactive sim. 👇 kevinxu02.github.io/real2sim-indoo…
Xingyi Yang
@yxy2168
Sep 22
📢 New survey: Efficiency in LLM Multi-Agent Systems 🤖 More agents can mean better results—but also more overhead. The key question: when is collaboration actually worth the cost? Our survey takes a closer look ↓ 🌐 lexiang-xiong.github.io/Awesome-Effici… ⭐ github.com/Lexiang-Xiong/…
00:00
1
Xingyi Yang
@yxy2168
Sep 15
I’ve always dreamed of something beyond backprop. Tried it for a while, but never quite made it work. Excited to see where this goes.
Sakana AI
@SakanaAILabs
Sep 14
Introducing PC-ALM, a local-learning alternative to backpropagation. Our method trains 1000-layer neural nets using only local dynamics, and without backprop. Blog: pub.sakana.ai/pc-alm/ Standard deep learning relies on backpropagation. The brain, however, cannot implement
00:00
1