Hidden Messages Between Diffusion Language Model Agents: Causal Evidence for Latent Communication

A research paper asks whether one independent diffusion language model trajectory can pass a compact hidden message into another, arguing that multi-agent systems typically discard the models' revisable continuous states and force agents to communicate only after decoding text. The work presents causal evidence for latent communication between diffusion language model agents, framing the approach as an alternative to text-only message passing.

Diffusion language models form answers through a sequence of revisable continuous states, but multi-agent systems usually discard those states and make agents communicate only after decoding text. We ask whether one independent diffusion trajectory can pass a compact hidden message into another and,