AI Reading Group — Diffusion Language Models

​The idea of Safe AI Lausanne’s reading group: we pick an article or research paper around AI or AI safety, we read and, and then we meet to discuss. It’s open to everyone interested in the topic!

​We start each session by clarifying together the points we didn’t understand in the paper, and then we reflect on the contributions of the paper, its implications, its strengths and weaknesses, or how its results could be taken further.

​This week, we read a new paper on Diffusion Language Models: Partition Generative Modeling: Masked Modeling Without Masks.

​Diffusion language models seem to be a promising direction that could significantly improve upon the current auto-regressive models, so we think it would be interesting to better understand how they work!

​Please take a quick look at the shared doc where you can see the plan for Thursday’s session, and write any questions you’d like to discuss, both in terms of understanding the paper, and in terms of reflection/discussion about it :)

​Then come on Thursday at EPFL to discuss it live!