Shital Shah Suggests Key-Preconditioned Distributions for AI Watermarking

Shital Shah, Principal Research Engineer at Microsoft Research AI Frontiers Lab, replied to discussion on steganographic methods for generative models. He stated one possibility is to generate an initial probability distribution already preconditioned on the key. The reply focuses on this approach to embedding information during generation. Among visible replies, Shah presents the preconditioning step as a potential method for handling keys in such systems. The post stands as his direct contribution to the exchange on these techniques for AI models. No further details or confirmations appear in the available posts.