Neel Nanda

This study investigates how circuit analysis techniques apply to a large language model.

Examining how AI interprets and interacts with the game Othello.

Activation patching reveals insights into language models' outputs and behaviors.

The study investigates universal neurons in GPT-2 models and their roles.

Researchers investigate how models adapt when components are removed.

A closer look at causal attribution methods for large language models.

Sparse autoencoders enhance the interpretability of AI systems and their decision-making processes.

Learn how transcoders help clarify complex language models.

This article examines how certain neurons affect uncertainty in language model predictions.

This study uses sparse autoencoders to interpret attention layer outputs in transformers.

JumpReLU SAEs improve data representation while keeping it simple and clear.

Gemma Scope offers tools for better understanding language models and improving AI safety.

New metrics improve understanding of Sparse Autoencoders in neural networks.

BatchTopK sparse autoencoders improve language processing through smart data selection.