Salman Rahman

I am a Computer Science Ph.D. student at the University of California, Los Angeles, working under the supervision of Professor Saadia Gabriel. My research focuses on AI alignment and improving the trustworthiness of artificial intelligence systems, particularly ensuring that future AI systems more advanced than humans remain aligned with human interests. For instance, as AI systems become capable of writing millions of lines of code or discovering novel knowledge beyond human expertise, how can we ensure their outputs remain safe and reliable?

My current research spans two main areas: Scalable Oversight of large language models (LLMs) and multimodal models, where I study methods to reliably evaluate LLM outputs on complex tasks that are challenging for direct human evaluation; and red-teaming of LLMs and multi-agent LLM systems to understand and mitigate potential risks. I collaborate with the NYU alignment group on these topics. Additionally, I am interested in the development of safe and reliable LLM agents and investigating the implications of LLM-assisted scientific research.

I interned at Apple's machine learning team in Summer 2024, where I worked on building, training, and evaluating efficient small-scale state-of-the-art multimodal LLMs for specialized computer vision tasks, designed for on-device deployment. At UCLA, I help organize the UCLA NLP Seminar Series - stay tuned!

During my undergraduate and master's studies, I focused on computational sustainability, exploring how AI systems can address pressing societal and environmental challenges of the 21st century. Explore my publications on computational sustainability here.