Sayan Ghosh

Hello!

I am currently a 4th year PhD Student at the University of Southern California.

These days, I am very interested in LLM evaluation, characterizing LLM behaviors by analyzing their longform text outputs, test-time scaling, and finding out whether we can better control the trade-off between the peaky distributions induced by alignment and the wilder, untamed distributions learnt by base models.

Previously, I was a research engineer at ASAPP. I studied Math and Computer Science at the University of Michigan, where I was fortunate to be advised by David Jurgens and Joyce Chai. I worked on several projects in areas ranging from pragmatics, sociolinguistics, network theory, and bias/fairness.

I am always willing to help and advise undergrad or masters students interested in NLP research. If you are interested in collaborating, please email me at ghoshsay@usc.edu!

Here is a current CV.