01
About
Hi! I'm Wei.
I'm a PhD student at Columbia.
I do philosophy of AI. I am fortunate to be advised by John Morrison. This summer I am working on a project on agentic moral alignment with Elizaveta Tennant.
Curriculum Vitae →
02
Publications
01
Fictionalism about Personas: Folk Psychology as an Interpretability Strategy
A fictionalist account of LLM mentality.
02
Building a Correct-by-Design Lakehouse for Humans and Agents
We show how typed table contracts, Git-like data versioning, and transactional pipeline execution can make illegal states unrepresentable and avoid partial effects.
03
Blog
Apr 2026
Sanity Checks for Distributed Alignment Search
DAS is one of the more promising methods in the causal interpretability toolkit. I ran three randomization-based sanity checks on the MoNLI experiment.
04
Contact
Email
ws2720@columbia.edu
GitHub
github.com/shengweiming