Research notes, and the occasional thought from life between Abu Dhabi, China and wherever the next conference is.
Notes from the 13th Heidelberg Laureate Forum (HLF 2026) — 200 young mathematicians and computer scientists, 20+ Fields, Turing and Nobel laureates, a global village, and a question about what any of it is for.

Large vision-language models describe images well but are vague about where things are. Perceptio makes the model emit segmentation and depth tokens first, then answer. Short note on the idea, the training tricks and the results.

I pulled every news mention, podcast, video, paper and social post about me (Chinese + English) into one place, and kept only the ones that could be verified as actually being about this Yuchen Li. Browse the full interactive index at /press/.