Talk: From Research to Real-World: Navigating the Challenges of AI in Emotionally-Engaging Dialog

Title

From Research to Real-World: Navigating the Challenges of AI in Emotionally-Engaging Dialog

 

Abstract

Recently, large-scale neural approaches to language and dialog have significantly pushed the frontier of what’s possible across research and industry. However, turning these into products that users love carries a whole set of different challenges. At Daimon Labs we train models for open-domain emotionally-focused dialog, from friendly chat, to venting, journaling and more. In this talk I will tour through some of our recent work: Training a small retrieval-oriented language model to perform as well as models 25x larger, techniques for running low-latency language models at scale, and the parallels and differences between academic AI research and the more applied research we do as a startup. I’ll also hope to provide a glimpse into our vision for the future and raise some questions about human computer interaction in a world full of AI agents.

 

Bio

Ryan Benmalek is the CEO & cofounder of Daimon Labs, a startup building deep dialog models focused on emotional intelligence. He received his Ph.D. at Cornell University, advised by Prof. Serge Belongie and Prof. Claire Cardie. His research interest is in computer vision and natural language processing. His research primarily focused on self-supervised learning, natural language generation, and active vision. Ryan was an NSF Fellow and conducted research at Apple Siri, Google Brain, Microsoft Research, and MILA.