Joshua Tint

Email: J.Tint@sms.ed.ac.uk

Website: https://jtint24.github.io/

Research keywords: Pragmatics, Fairness evaluation, Sociotechnical issues

Bio:

Joshua an NLP researcher whose interests lie in evaluating and improving social communication in LLMs. He is originally from Arizona where he got a B.S. and M.S. in computer science from Arizona State University as a Flinn scholar. While there, he completed a baccalaureate thesis on understanding LLM responses to queer slang. He also served as a visiting research assistant at LMU Munich, where he investigated interpretations of culturally-specific gestures in VLMs. Prior to joining the University of Edinburgh, he was an engineer at Amazon.

In addition to his academic work, Joshua has also volunteered with Queer in AI and the Free Software Foundation.

PhD research:

LLMs have excelled at tasks with quantitatively verifiable rewards, but still often misunderstand context, miss social cues, behave unnaturally, and exhibit bias. In order to be deployed responsibly, language models must develop a better social understanding, which requires novel evaluation methods that reckon with social nuance. Joshua’s research focuses on building those evaluation systems, with a focus on improving in-context and real-world evaluation, developing methods that capture implicit and socially situated forms of communication, and understanding LLM output rhetorically.

Supervisors: Adam Lopez, Björn Ross