Giang Nguyen

• Training: 8B interpretable language models pretrained on trillions of tokens. [1] [2]
• Explanation: Attribution methods for model behavior at the concept, prompt, and input-to-concept levels. [1]
• Control: Controllable generation for topic control and safety alignment without fine-tuning. [3] [4]

Research on interpretable LLMs for tabular data.

Thesis: Transforming the black-box decision-making of AI models into explain-then-answer processes
Advisor: Anh Nguyen · Presidential Graduate Research Fellow

Thesis: Overcoming catastrophic forgetting by deep visualization

A position paper arguing that feature attribution in generative language models is under-specified: the same attribution method can answer different questions depending on what you are explaining. Introduces the Attribution Contract, a framework for naming the explanatory setting under which a feature-attribution claim is made.

The first large-scale language model with interpretability built into the architecture: every generated token traces back to input context, training data, and human-understandable concepts. Self-monitors for memorized content and suppresses it at inference time without retraining. Featured in TechCrunch.
[blog] [github] [huggingface] [techcrunch]