Leo Tronchon

Hugo Laurençon and Leo Tronchon from Hugging Face discuss their cutting-edge work on multimodal models like IDEFICS and OBELICS. They dive into the evolution of multimodal training, sharing challenges related to data quality and the intricacies of processing raw HTML. The conversation highlights the importance of image resolution for OCR and the hurdles faced in video data processing. Both researchers express optimism for open-source models, aiming for enhanced performance while tackling issues like hallucinations. Their insights reveal a bright future for multimodal AI innovation.