Dr. Max Bartolo (Presenter)

Max Bartolo is a Senior Research Manager at Cohere, where he recently led post-training for Command R+, the company’s flagship model and one of TIME’s Best Inventions of 2024. He also chairs the Data-centric Machine Learning Research (DMLR) working group at MLCommons, shaping best practices for large-scale model training. Max’s PhD focused on the adversarial robustness of language models under the supervision of Professors Pontus Stenetorp and Sebastian Riedel at University College London (UCL). Previously, he conducted research at Google DeepMind, Facebook AI Research (FAIR) and Bloomsbury AI. As an Adjunct Teaching Fellow at UCL, he developed and taught the Natural Language Processing module at the School of Management (SoM). Max’s research has won multiple awards, including Outstanding Paper awards at ACL ’22 and EMNLP ’24, and a Best Paper award at NeurIPS ’24. His work has been featured in VentureBeat, Wired, Fortune, and MIT Technology Review. Passionate about bridging cutting-edge LLM research with real-world impact, Max sits on the advisory boards of multiple AI startups applying advancements in model robustness and alignment to real-world challenges.

In this talk, Max will dive into the training methodologies that power cutting-edge conversational AI systems like ChatGPT. He will unpack the key techniques involved, including pre-training, Supervised Fine-Tuning (SFT), and Learning from Human Feedback (LHF). Following this, Max will lead an interactive, hands-on session where participants will learn how to train a small-scale Generative Pretrained Transformer (GPT) model using the Transformers library. This session will help you develop an intuitive understanding of how these AI systems learn from text data to generate intelligent responses.