We are excited to continue our series with Max Bartolo (Google DeepMind) who will be talking about Context: From Tokens to Capabilities.
The talk will be hybrid and you will be able to listen and ask questions both in person or remotely. After the talk there will be a catered networking session. Please note that the link to the session will be available at the end of the confirmation email, where it says "Additional information".
Abstract
This talk discusses the evolution of considerations around context in LLMs, including selected highlights from the speaker's work. It traces the journey from early question answering systems to advanced retrieval-augmented deep learning and modern LLMs. Highlights include the challenges with human feedback, procedural knowledge in pretraining, and novel methods for improving model robustness, including prompt sensitivity, the detection of under-trained tokens, the concept of implicit self-correction in-context, and RL for reverse engineering model context.
About the speaker
Max is a researcher at Google DeepMind and co-chair of the DMLR working group at MLCommons, shaping best practices for large-scale model training. His research focuses on language model robustness, complex reasoning, and innovations in dynamic adversarial data generation and benchmarking. He previously led the Command (post-training) modelling team at Cohere, worked at Facebook AI Research and Bloomsbury AI, and was also an adjunct teaching fellow at University College London. His research has been featured in leading global publications, including Wired, Fortune, and MIT Technology Review, has been recognised as one of TIME's best inventions of 2024, and has earned multiple awards at top-tier conferences.
Sponsors
This event is co-hosted by DocSoc and sponsored by Iconic.
