Christopher Potts

A dataset tests language models on self-referential language tasks.

Research shows GPT-2 struggles with languages that defy human patterns.

Explore how interpretability illusions affect our view of neural networks.

A study on how language models process and understand complex linguistic tasks.

A study assessing various methods for interpreting language model neurons.

MoEUT improves Universal Transformers' efficiency and performance in language tasks.

A detailed study on how models memorize text and its implications.

Innovative methods enhance LLMs alignment with human preferences for better performance.

An in-depth look at Gated Recurrent Units in sequence learning.

This approach simplifies choosing effective pretraining datasets for language models.

A new dataset enhances query autocomplete suggestions using real user data.