Latent multi-agent LLM accuracy gains decompose into content-independent and example-specific parts that can oppose each other, so end-task scores alone do not prove genuine latent communication.
Enabling Agents to Communicate Entirely in Latent Space , booktitle =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Do Latent Channels Actually Communicate? A Causal Audit of Latent Multi-Agent LLM
Latent multi-agent LLM accuracy gains decompose into content-independent and example-specific parts that can oppose each other, so end-task scores alone do not prove genuine latent communication.