Deema Alnuhait

I am Deema Alnuhait, a second-year computer science Ph.D. student at the University of Illinois Urbana-Champaign (UIUC) , advised by Prof. Hao Peng. I am also a visiting student at Argonne National Laboratory and previously interned at Amazon.
Before this, I completed my master’s degree at Columbia University, where I worked with Prof. Zhou Yu.

I am looking for research internship opportunities for 2025 summer

Email  /  CV  /  Scholar  /  Twitter  /  Github

Research

My research focuses on enhancing the robustness and trustworthiness of large language models (LLMs), including hallucination detection, bias and toxicity mitigation, and model alignment.

FactCheckmate, a framework for preemptively detecting and mitigating hallucinations in LLMs by analyzing hidden states to identify issues before they appear in outputs, using effective interventions with minimal overhead to improve factuality.

AraTrust, a comprehensive Arabic-specific trustworthiness benchmark for LLMs, covering categories such as ethics, safety, and offensive language. It evaluates various models, demonstrating GPT-4's superior performance over open-source alternatives like AceGPT and Jais.

CIDAR, an open Arabic instruction-tuning dataset developed with extensive manual review for cultural and linguistic alignment, addressing biases in machine-translated datasets. Models fine-tuned on CIDAR demonstrate improved cultural relevance and performance in Arabic-specific tasks compared to those fine-tuned on larger but less tailored datasets.

Education

Internship Experience

Copyright © Jon Barron