🛬 Attending

🇰🇷ICML2026, Seoul, KR

My research is supported by:

Coefficient Giving

TML

Zulip

<aside> <img src="notion://custom_emoji/0d39d0ab-438c-4f29-be70-03aa9d912057/2c928fcd-40c2-806f-bf66-007a2758f01d" alt="notion://custom_emoji/0d39d0ab-438c-4f29-be70-03aa9d912057/2c928fcd-40c2-806f-bf66-007a2758f01d" width="40px" /> Google Scholar

</aside>

<aside> <img src="notion://custom_emoji/0d39d0ab-438c-4f29-be70-03aa9d912057/2c928fcd-40c2-8076-afbe-007af223fbba" alt="notion://custom_emoji/0d39d0ab-438c-4f29-be70-03aa9d912057/2c928fcd-40c2-8076-afbe-007af223fbba" width="40px" /> Research Gate

</aside>

<aside> <img src="attachment:5a1e77b6-a69a-4a9c-8282-b0e70b4de887:Less_Wrong_LOGO.png" alt="attachment:5a1e77b6-a69a-4a9c-8282-b0e70b4de887:Less_Wrong_LOGO.png" width="40px" /> LessWrong

</aside>

Socials

<aside> <img src="notion://custom_emoji/0d39d0ab-438c-4f29-be70-03aa9d912057/2c928fcd-40c2-8012-8d55-007a1a6ff476" alt="notion://custom_emoji/0d39d0ab-438c-4f29-be70-03aa9d912057/2c928fcd-40c2-8012-8d55-007a1a6ff476" width="40px" />

Linkedin

</aside>

I work on AI trust for (multi-) agentic system that communicates, reasons and acts autonomously, making such system smarter (reasoning) and safer (alignment) at the same time. My reasoning work primarily focus on multimodal physics-physical reasoning with the applications in AI for Physical Science and Embodied Intelligence. My safety work revolves around spontaneous misalignment (such as sycophancy, deception and collusion).

My vertical focus on reliability and generalizability of evaluation and post-training, where I often leverage horizontal methods from interpretability and formal verification in Lean:

• Evaluation: The shift from static → dynamic agentic evaluation calls for more study into their robustness/faithfulness: I lead a MARS stream on eval hacking that study how AI interaction with different aspects of eval could compromise eval:
  • interact with data: contamination, reasoning-memorization interplay in MLLMs
  • interact with system: proxy/metric gaming, inverse RL of reward function
  • interact with agents: collusion, persuasion and herding effect in multi-agent
• Post-training: Cross-domain transfer/generalization of reasoning capability & reliability
  • Transfer across modality: how does visual-textual info interplay during RL (SeePhys/SeePhys Pro)
  • Transfer across domain: how do different domain-specific experts interfere or synergize each other in post-training (such as MOPD in DeepSeek-V4)?
  • Transfer across training paradigm: when do SFT/RL/OPD work well and can we interpret such dynamic from weight space?

Useful stuff I (personally) recommend:

• I use Ferdium to aggregate all of my notifications in one place, such that I only need to check 1 app when I wake up, highly recommend! I heard Hermit does a similar job on mobile device but haven’t gotten to try it yet.
• I use OpenAI Codex for all my coding now, which is currently the best AI coding tool out there in my opinion (and it’s open-sourced unlike the other one)
• OpenAI RFT (the finetuning API) doesn’t work, it’s expensive and will fail with no reason
• ai2-asta for literature review NO hallucination on Semantic Scholar backend
• uv for package management, don’t use conda
• inspect eval
• LMMS-eval for multimodal eval
• Tinker: I’m a beta user of the Tinker API
• slime: RL Scaling with Megatron-LM for training + SGLang for inference