William Rudman

@WilliamRudmanjr

Postdoctoral Fellow @UTAustin. PhD in Computer Science from @BrownUniversity. Interested in VLM/LLM (mechanistic) interpretability and efficiency.

Austin, TX

wrudman.github.io

Joined December 2020

131 Following

171 Followers

Replies Reposts Media

Pinned
William Rudman
@WilliamRudmanjr
Jan 14
Why do vision–language models follow incorrect prompts even when the image clearly disagrees? We study this behavior as prompt-induced hallucination (PIH) in object counting. [1/6]
1
William Rudman
@WilliamRudmanjr
Oct 6
Can models predict their own behavior? Check out our #COLM2026 paper (poster presentation today at 4:30pm) showing that even SOTA VLMs contradict their stated introspective rules for “when to call an apple red.”
Michal Golovanevsky and 3 others
1
William Rudman
@WilliamRudmanjr
Jul 6
@MichalGolov and I will be presenting our work locating causal mechanisms of “prompt induced hallucinations” in VLMs at #ACL2026 tomorrow at 11am in the Grand Hall! 📖: arxiv.org/pdf/2601.05201
Dana Arad
William Rudman
@WilliamRudmanjr
Jun 17
Hypothesis -> experiments -> analysis -> conclusions. LLMs are great at writing code and conducting experiments. But there’s a weakness in their ability to propose statistical models and evaluate their fit. Enter VESTA: Visual Exploration with Statistical Tool Agents.
Jessy Li and 5 others
1
William Rudman
@WilliamRudmanjr
Nov 11, 2025
Do you hate running Path Patching on larger models? Accelerated Path Patching can save days of compute time. APP produces faithful circuits by using contrastive pruning to reduce the search space of circuit discovery methods. Go save some trees 🌳by using APP to speed up PP.
Ruochen Zhang
@ruochenz_
Nov 11, 2025
⚠️Circuit-finding too slow? We introduce Accelerated Path Patching (APP), a method that uses task-specific pruning to cut down the search space before patching. 🚀APP finds faithful circuits with up to ~93% faster runtime.

William Rudman

William Rudman’s posts