Hi! 👋
I'm Alexander Wan, an undergraduate at UC Berkeley majoring in Computer Science. I'm broadly interested in machine learning and NLP, particularly in building better model evaluations. I'm currently doing research at Stanford CRFM. I formerly worked on LLM robustness & security at the Berkeley NLP group, and interned at the MSU Heterogeneous Learning and Reasoning lab.
See my: LinkedIn / Github / Google Scholar / Twitter
News
Nov 2023
I gave a talk at USC ISI's Natural Language seminar on the manipulation of LLMs through data.
Apr 2023
Our paper on poisoning instruction-tuned models was accepted to ICML.
Publications
Stanford Response to the US AI Safety Institute Request for Comment on Misuse of Dual-Use Foundation Models
Rishi Bommasani, Alexander Wan, Yifan Mai, Percy Liang, Daniel E. Ho
What Evidence Do Language Models Find Convincing?
Alexander Wan, Eric Wallace, Dan Klein
ACL 2024 (Main)
Poisoning Language Models During Instruction Tuning
Alexander Wan*, Eric Wallace*, Sheng Shen, Dan Klein
ICML 2023
GLUECons: A Generic Benchmark for Learning Under Constraints
Hossein Rajaby Faghihi, Aliakbar Nafar, Chen Zheng, Roshanak Mirzaee, Yue Zhang, Andrzej Uszok, Alexander Wan, Tanawan Premsri, Dan Roth, Parisa Kordjamshidi
AAAI 2023
Projects
DIY infini-gram
- Prior to the official code-release, built an open-source implementation of infini-gram, a method for analyzing and integrating LLMs with extremely large text corpora using suffix arrays. (May 2024)
- Built upon this work by replacing the suffix array with an FM-index and wavelet trees to reduce disk-usage by 7.5x. (June 2024)
Miscellaneous
-
I was an instructor at InspiritAI, where I introduced AI concepts and Scratch programming to 5th/6th graders.
-
I'm a member of Machine Learning @ Berkeley.
-
I am occasionally active on the Artificial Intelligence StackExchange, answering questions about AI.
Contact
Email: first 4 letters of first name + last name [at] berkeley [dot] edu