This video explains an architecture Yoshua Bengio is excited by: generative flow networks, or GFlowNets. The idea behind GFlowNets is to represent the way humans reason sequentially, adding a new piece of relevant information at each step. It consists of both RL since the model learns a policy to solve a problem and generative modeling because it generates trajectories to get to its end state in the graph network.

Paper for more technical information: https://arxiv.org/pdf/2106.04399.pdf
Yoshua's Notion page on this: https://milayb.notion.site/The-GFlowN...
My high-level notes: https://shizacharania.notion.site/GFl...