Haoyu Zhao

🎓 Research Fellow @NUSingapore Computing, LV-Lab | PhD Fudan University | 🤖 Embodied AI · 🌍 World Models · ✨ Generative AI

[Haoyu Zhao @zhaohaoyuu]

@XueJia24682 Relying only on 2D vision or using 3D vision? Further, I think the robot with four wheels moves more flexibly than a human in this case.

[Haoyu Zhao @zhaohaoyuu]

@Artedeingenio 👍👍👍

[Haoyu Zhao @zhaohaoyuu]

@cherry_mx_reds To be honest, I really like this.

[Haoyu Zhao @zhaohaoyuu]

Amazing! GPT-6 Astra directly generated this animation using JavaScript and Canvas. 🐢🐇

The video brings the classic fable “The Tortoise and the Hare” to life—entirely through code-generated visuals.

[Haoyu Zhao @zhaohaoyuu]

@scottstts @thsottiaux Awesome application!

[Haoyu Zhao @zhaohaoyuu]

@cherry_mx_reds Regeneration using the given prompt via GPT-6 Astra, but obviously Opus performs better.

[Haoyu Zhao @zhaohaoyuu]

@TheMathFlow Still work for a high-order cube?

[Haoyu Zhao @zhaohaoyuu]

🚀 Can MiniMax-H3 reason about the physical world?

We build 517 cases across 4 scenarios to probe omni-modal physical-world reasoning. H3 achieves 41.97% overall success—promising, but multimodal integration remains challenging.

🤗 https://t.co/10x2KFNK9I

[Haoyu Zhao @zhaohaoyuu]

@Kuvvius Impressive spatial intelligence from GPT-6.

[Haoyu Zhao @zhaohaoyuu]

Interesting finding from our GPT-6 robot experiments 🤖

After each trial, we let the model summarize its experience and carry it into the next run. With this context, execution became smoother—and the robot completed the task in one shot.

Experience as context. See the video.