Xiang Fu
Scholar

Xiang Fu

Google Scholar ID: 3CzDreAAAAAJ
Boston University; Modularium Research
Homepage↗Google Scholar↗
Citations & Impact
All-time
Citations
11
 
H-index
2
 
i10-index
0
 
Publications
4
 
Co-authors
0
 
Contact
CVOpen ↗GitHubOpen ↗LinkedInOpen ↗
Publications
1 items
LLMs Learn Better In-Context from Rules than from Examples
2026
Cited
0
Resume (English only)
Academic Achievements
  • Paper "Who’s the Impostor? Multi-Agent Social Deduction for Evaluating LLM Social Reasoning" accepted to NeurIPS 2025 Workshop on Evaluating the Evolving LLM Lifecycle
  • Collaborative work "Humans have more sophisticated strategies for processing semantically anomalous word pairs than do Large Language Models" to appear at the 66th Annual Meeting of the Psychonomic Society (with Catherine Caldwell-Harris)
  • Co-authored paper "M^2IV: Towards Efficient and Fine-grained Multimodal In-Context Learning via Representation Engineering" accepted to COLM
  • Preprint published on arXiv in May 2025: "Can an Easy-to-Hard Curriculum Make Reasoning Emerge in Small Language Models? Evidence from a Four-Stage Curriculum on GPT-2" (arXiv:2505.11643 [cs.CL])
  • Awarded a $6,000 Student Research Award (SRA) from Boston University’s Undergraduate Research Opportunities Program (April 2025)
  • Presented a poster at Boston University’s 28th Annual Undergraduate Research Symposium (October 2025)
Background
  • Third-year undergraduate student at Boston University's Faculty of Computing and Data Sciences (CDS), majoring in Data Science and Computer Science
  • Research interests lie at the intersection of language model behavior, evaluation, and alignment
  • Aims to understand internal mechanisms of how language models represent and transform information
  • Designs metrics to reveal model strengths and blind spots
  • Develops alignment techniques to steer model outputs toward human values and intent
  • Explores the interface between machine and human cognition, examining how artificial models generalize, reason, and communicate compared to humans
  • Founded Modularium Research to study model behavior and improve evaluation methodologies
Miscellany
  • Maintains a personal course resource website: BU Courses (https://bucourses.com)
  • Writes essays and reflections on language processing, cognition, and AI
  • Interests include photography, writing, cartography, and creative experimentation
  • Philosophy: "Train my eyes instead of inheriting the world’s well-trodden gaze"
  • Shares PGP public key, emphasizing digital privacy and security
Co-authors
0 total
Co-authors: 0 (list not available)