Questions for the seminar paper "GAIA-2: A Controllable Multi-View Generative World Model for Autonomous Driving".
-----------------------------------------------------------------------------------------------
Please send your answers to: galessos@cs.uni-freiburg.de

1) How does GAIA-2 allow for conditional generation based on natural language prompts, without training on such modality? (2 sentences)
2) What is the peculiarity of the distribution of "tau" used for training GAIA-2, and what is its purpose? (2 sentences)
3) What do you think of the experimental setup of the paper? Do you find the presented results satisfactory? (1 sentence)