Projects

Learn a powerful motion prior with diffusion models, and use it to denoise, restore, and imagine better motion!

CS194-280 Advanced LLM Agents project. One open question in the study of Large Language Models (LLMs) is whether they can emulate human ethical reasoning and act as believable proxies for human judgment. To investigate this, we introduce a benchmark dataset comprising 196 real-world ethical dilemmas and expert opinions, each segmented into five structured components. We also collect non-expert human responses for comparison, limited to the Key Factors section due to their brevity. We evaluate multiple frontier LLMs using a composite metric framework based on BLEU, Damerau-Levenshtein distance, TF-IDF cosine similarity, and Universal Sentence Encoder similarity. Metric weights are computed through an inversion-based ranking alignment and pairwise AHP analysis, enabling fine-grained comparison of model outputs to expert responses.

This is a fundamental track project for COMPSCI194-196 LLM Agents and LLM Agents hackathon. We focused on exploring robust and generalizable internal representations of lightweight LLMs and investigating the progression of learned features with linear probes and sparse autoencoders in OthelloGPT. Our experiments reveal that SAEs provide a more robust and disentangled decoding of the features the model is learning, particularly for compositional attributes.

This is a Meta Quest track project for the Stanford XR Hackathon. We focused on recovering human psychological dysfunctions, aiming to provide a comprehensive treatment protocol by designing multiple simple interactive meditation games utilizing the power of Meta Quest3. We build interactive environments from scratch in Unity; users can choose different environments, background music, and meditation guidance in Zen.