Every month, the joint laboratory invites speakers to take part in seminars for its partners.
Mathieu Labeau (Télécom-Paris): “From GPT-3 to ChatGPT: integrating human feedback in Language Models”
Abstract:In this presentation, we will focus on the paper Training language models to follow instructions with human feedback (Ouyang et al, 2022) – NeurIPS 2022; the goal being to align a language model to its user’s intent. We will come back on some background on language models: the capabilities of pre-trained LMs, and those improved with supervised fine-tuning. Then, we will detail the paper’s contribution, being a large scale and generalist application of reinforcement learning from human feedback (RLHF) to language models.
Bio:Matthieu Labeau joined Télécom Paris as a Senior Lecturer in 2019. He completed his doctorate at the University of Paris-Saclay, and became interested in the problems posed by large vocabularies in language modeling. He then became a postdoctoral fellow at the University of Edinburgh. His research areas, within automatic language processing, mainly concern representation learning and language modeling.
Links:
- https://www.telecom-paris.fr/matthieu-labeau?l=en
