Haoxiang Sun 孙浩翔

Hello! I am a 4th-year undergrad in CS and Finance at Renmin University of China. My research focuses on Language Models. I am fortunate to work with Sewon Min at UC Berkeley, and previously with Manling Li and Zihan Wang (Northwestern) and Xin Zhao (Renmin). I am applying to CS PhD programs for Fall 2027.

As a visiting student at Berkeley, I took CS 288: Advanced NLP (A+, highest among ~150 students, taught by Sewon Min and Alane Suhr) and CS 294-258: Computational Models of Language Use (A+, taught by Alane Suhr). I have also been awarded the SenseTime Scholarship and two National Scholarships, and in high school, I won a silver medal at the China Mathematical Olympiad (CMO).

Publications & Preprints

Selected Awards

  1. 2026 Second-Class Academic Scholarship Renmin University of ChinaCNY 3,000
  2. 2026 CCF Elite Collegiate Award 103 awardees nationwide
  3. 2026 SenseTime Scholarship 30 awardees nationwideCNY 20,000
  4. 2024, 2025 National Scholarship (×2) Ministry of Education, top 0.4% nationwideCNY 10,000 each
  5. 2024 Asia Team Champion Asia Terminal, hosted by Citadel | Citadel SecuritiesUSD 1,500
  6. 2022 Silver Medal, China Mathematical Olympiad (CMO) 2nd in the Anhui team at the final
  7. 2021, 2022 First Prize (×2), Chinese High School Mathematics League Anhui Province
  8. 2021 Gold Medal, China Southeast Mathematical Olympiad

Teaching & Services

  1. 2026 Reviewer, ATTRIB Workshop @ NeurIPS 2026
  2. 2026 Teaching Assistant, BAIR-NCKU Summer Program RAG course taught by Sewon Min

Invited Talks

  1. Sep 4, 2026 Models All the Way Down: Auditing the Invisible Dependencies of Modern LLMs Slides Open Ecosystem Team, Allen Institute for AI (w/ Sanjay Adhikesaven, P1-41)
  2. Jul 7, 2026 OlymMATH: An Olympiad-Level Math Benchmark for LLMs Slides The 64th Annual Meeting of the Association for Computational Linguistics
  3. Jun 8, 2026 ModSleuth Demonstration Slides Open Ecosystem Demo Day, Allen Institute for AI

Blog Posts

  1. Sep 8, 2026 Is My AI Model American or Chinese? Tracing the Nationality of Model Dependencies Research A nationality census of ModSleuth's dependency graph. In seven American open releases, half the models that generated the training data were released in China, because from late 2024 the strongest models anyone could download and reuse freely were Chinese. The label names the builder; the lineage names whoever led open AI when the model was built.

Life

One of my favorite things about living in the Bay Area is exploring on foot and by public transit (BART, Caltrain, Muni, and more). A go-to adventure of mine is taking BART to Powell Street Station, walking through Chinatown up to Fisherman’s Wharf, and then strolling along the waterfront all the way back to Embarcadero Station. I also love riding buses across the Richmond-San Rafael Bridge, the Golden Gate Bridge, and the Bay Bridge. Another frequent trip of mine is taking Caltrain from San Francisco all the way down to San Jose, grabbing a seat on the upper deck and enjoying the scenery along the way. I’m also a big fan of the vibe around Ocean Beach in San Francisco. When I’m not out exploring the Bay Area, I also enjoy playing Minecraft, especially on the Hypixel server, where my favorite game modes are Capture the Wool and Bed Wars (particularly Rush and Castle 40v40).

Flight Map

Friend Links

It seems to be a tradition to keep some links here, and I am happy to carry it on. Many of the people here are researchers I deeply admire, and I have been lucky to learn from them as collaborators, mentors, and friends. The order shuffles on every refresh, and the list is always open. Just let me know if you’d like to be here too.

Jiajun Liu Yihan Yang Sanjay Adhikesaven Stephen Xie Yuheng Wang Yuxia Ding Zihan Wang Ryan Wang Prasann Singhal