Weekly Seminar - Kolby Nottingham

About this event

Abstract Modern large language models are optimized to be helpful and harmless, but strict alignment suppresses the creative entropy of the underlying pretrained model. As progress drives SOTA to astonishing capabilities, it also strips the spontaneity needed for dynamic storytelling. This talk explores the tensions between grounded and unhinged narrative. We will dive into training techniques, feedback signals, and system design required to transform modern LLMs into compelling, adaptive interactive storytellers. Biography Kolby Nottingham is the Head of Al at Latitude, leading the Al and Game Engine teams in building real-time generative Al gameplay for the RPG platforms AI Dungeon and Voyage. He holds a PhD in Computer Science from UC Irvine where his research focused on applying reinforcement learning and language modeling for grounding LLM agents in gaming, simulation, and embodied environments. He has publications across ICML, NeurIPS, and NAACL venues.