Consistency Training Along the Transformer Stack
attention and MLP-based consistency training techniques across persona, jailbreak, sycophancy, and prefill attacks
View projectHello! I'm
Caroline Wei
Computer Science + Math @ UCLA
Alignment Research
I am a 4th year Computer Science student at UCLA interested in all things cool and exciting. My current interests lie in alignment research, protein language models, forecasting and modeling, coffee, mahjong, boxing, and periodically updating my Substack.
Fun Stuff
Federated Learning for Unlearning
exploring secure unlearning methods in federated learning settings for data privacy
View projectExperience
-
09.26 – now
Research Assistant, UCLA Security Lab
Reward hacking and agent trajectory analysis
-
06.26 – 08.26
Security, Risk, Compliance Intern, Gilead Sciences Inc.
NLP-based resume parser to automate CV process
-
03.26 – 06.26
Research Fellow, SPAR
Exploring consistency training methods
-
09.25 – 02.26
Data Engineer (Contract), DENSO Robotics
Forecasting model for OOD prediction on product demand