Baicheng Chen

CSE undergrad @cuhksz. Working on #TrustworthyAI.

27 Fall PhD applicant.

[Baicheng Chen @dannychen1223]

Today’s LLM agents can write code, call tools, and debug programs.

But can they reverse-engineer a cryptographic binary end to end?

We built CREBench to find out.

Thrilled that our work has been accepted to @COLM_conf 2026! 🎉

🧵👇

#COLM2026 #AI4Security #Agent #RE

[Baicheng Chen @dannychen1223]

9/

Paper: https://t.co/tQHQ8UXkOd
Code & Benchmark: https://t.co/lVDaV7PA4Y

Huge thanks to all the incredible collaborators and mentors who made this possible: Yu Wang, Ziheng Zhou, Xiangru Liu, Juanru Li, and Tianxing He!

I'm happy to chat and see you in San Francisco! 🌉

[Baicheng Chen @dannychen1223]

Today’s LLM agents can write code, call tools, and debug programs.

But can they reverse-engineer a cryptographic binary end to end?

We built CREBench to find out.

Thrilled that our work has been accepted to @COLM_conf 2026! 🎉

🧵👇

#COLM2026 #AI4Security #Agent #RE

[Baicheng Chen @dannychen1223]

8/

I also genuinely appreciated the @COLM_conf review process.

One reviewer wrote that they “enjoyed this paper a lot,” and the meta-review called CREBench “a model paper for how agentic benchmarks should be discussed and released”.

This meant a lot to us.

[Furong Huang @furongh]

Hot take: robots should not dream in pixels.

Pixels are too low-level.
Latents are too opaque.

μ₀ predicts a third thing:
3D motion traces.

On real robots, it beats π₀.₅ — with ~1/100 the data scale and no action labels for world-model pretraining. 🧵

https://t.co/UfmrqNlBtw

(this video features voiceover narration)