I'm broadly interested in studying the science of deep neural networks, generalization, and learning theory. I'm also very interested in AI alignment and human welfare over the next century.

I'm currently working on post-training video-to-video models at Decart, and have previously studied subliminal learning in VLMs, the Platonic Representation Hypothesis, and representational alignment between discrete diffusion and autoregressive models.

Some stuff I think is interesting :)

• https://arxiv.org/pdf/2604.19740
• https://arxiv.org/pdf/2507.14805
• https://x.com/juli_li_/status/2066210558630863352
• https://safe.ai/ai-risk
• https://arxiv.org/pdf/1612.00796