Siqiao Huang (@knightnemo_) 's Twitter Profile
Siqiao Huang

@knightnemo_

Undergrad, Yao Class @Tsinghua_Uni . Interested in World Models, Generative AI and Reinforcement Learning.

ID: 1823740396876570624

calendar_today14-08-2024 15:16:16

53 Tweet

30 Followers

598 Following

Tim Rocktäschel (@_rockt) 's Twitter Profile Photo

World Models already had their "Rich Sutton can't fit into the workshop room" moment! AI is truly accelerating at an immense speed. x.com/pabbeel/status…

Siqiao Huang (@knightnemo_) 's Twitter Profile Photo

Thank AK for sharing Vid2World, our work on crafting video diffusion models to interactive world models! Project Page: knightnemo.github.io/vid2world/

Yitang Li (@li_yitang) 's Twitter Profile Photo

🤖Can a humanoid robot carry a full cup of beer without spilling while walking 🍺? Hold My Beer ! Introducing Hold My Beer🍺: Learning Gentle Humanoid Locomotion and End-Effector Stabilization Control Project: lecar-lab.github.io/SoFTA/ See more details below👇

MikaStars★ (@mikastars39) 's Twitter Profile Photo

As a researcher who has also been a bachelor top-conference publisher, there are three points I personally believed to be true: 1. generally the time anyone firstly published a paper is not a big deal (could be second year in phd or first year in bachelor). the quality matters.

Google DeepMind (@googledeepmind) 's Twitter Profile Photo

What if you could not only watch a generated video, but explore it too? 🌐 Genie 3 is our groundbreaking world model that creates interactive, playable environments from a single text prompt. From photorealistic landscapes to fantasy realms, the possibilities are endless. 🧵

Siqiao Huang (@knightnemo_) 's Twitter Profile Photo

Totally agree, Albert Gu. In the end, we want models with inductive biases that match its tasks. One special note on transformers: If you want to get transformer's inductive bias, you should probably look at areas that transformers hasn't worked that well yet.🤫

Siqiao Huang (@knightnemo_) 's Twitter Profile Photo

My Predictions on world models: 1. They will help robot learning, but not every task. Neural game engines will come first 2. Physics will be solved by data scaling, no need for specific design. 3. Consistent Memory needs ssm-style hidden state, but it's a relatively small prob

Siqiao Huang (@knightnemo_) 's Twitter Profile Photo

I want to quietly mention that we basicly came up with the same idea of code execution as world models six months ago, and it turned out to be a EMNLP'25 paper (top 0.5% meta score). Check out: arxiv.org/abs/2502.11167 . Glad to see META pushing it a lot further.