We investigate GPT-4’s ability to identify harmful gender, racial, and class biases within written literature, as well as its own generated stories. We find that GPT-4 does not achieve a high enough accuracy to be able to de-bias narratives without human oversight.
Body imitation by a humanoid simulation using Azure Kinect SDK for body tracking, ROS, and the Darwin-OP. The system can process a video of human body movements, categorize the movements, and then accurately imitate it on a humanoid robot.
Website that connects users to charitable organizations providing relief for natural disasters around the world.
This project presents a development process to improve upon Boston Dynamics's quadruped robot Spot navigation behaviors to produce behaviors that are more dog-like in appearance.
Winner of 2022 CNS Award for Excellence in Computer Science and Computer Engineering. Published HICSS 2024.
Body imitation by a humanoid simulation using Azure Kinect SDK for body tracking, ROS, and the Darwin-OP. The system can process a video of human body movements, categorize the movements, and then accurately imitate it on a humanoid robot.
Website that connects users to charitable organizations providing relief for natural disasters around the world.
Trained on the Austin Crime database entries over the past year and uses multiple different models to try and predict whether a crime will be cleared or not by the police department in the district.
Making an ELECTRA-small model trained on the large-scale labeled Stanford Question-Answering Dataset robust against an adversarial challenge set produced by Robin Jia and Percy Liang (2017) through fine tuning and inoculation experiments.
Reinforcement learning-based pandemic simulator model that models the UT Austin population of students and staff, as well as university-specific locations such as apartments, campus, and parties.
