Unclaimed profileBuilt from public GitHub and Hugging Face activity. Feifan has not joined Lounge yet.Claim it
Not on Lounge yet
Feifan Liu
developerBusiness Honors + Computer Science @ UT Austin |
AI safety + interpretability
Public projects
llm-epistemic-uncertainty
Code and experiments for "Decodability and Causal Control of Epistemic Uncertainty in LLMs" preprint
Pythonactivation-steeringinterpretabilityllm
ARENA
Deep learning fundamentals: transformers, interpretability, RL, LLM evals + alignment
Jupyter Notebook
reverse-engineering-grokking
A foray into mechanistic interpretability using the case of a 1-layer transformer grokking modular addition -- techniques include principal component analysis, discrete Fourier transforms, ablation, linear probing, and activation patching
Jupyter Notebook
neuralnetwork
Constructing a neural network without PyTorch to learn how to implement backpropagation, gradient descent, and learn more about model architecture.
Jupyter Notebook
Likely stack
PyTorchTransformers
