Questions for the seminar paper "Diffusion model as world model"
-----------------------------------------------------------------------------------------------
Please send your answers to: sahayr@cs.uni-freiburg.de

1) Describe auto-regressive drift and how the authors get over it. (1-2 sentences)
2) What is the main limitation of training the model using random trajectories? (1-2 sentences)
3) What are the drawbacks of the proposed paradigm shift from 'games as code' to 'games as weights in a neural network'? (3 sentences)