J-Comm links LLM agents with hidden and Jacobian latent channels
Qwen3-8B to Qwen3-4B communication scored highest across seven benchmarks and cut generated text tokens by 19.2–74.8% versus TextMAS.
Stories per day · 30 days
Relatedmulti-agent2benchmarks1context-compression1cross-attention1diffusion-language-models1hidden-messages1jacobian1qasc1question-answering1qwen1
7 stories, newest first · one card per story
Stories tagged #latent-communication
Diffusion LM agents pass usable hidden messages in latent states
On QASC, matched messages reach 80.8%, while a wrong-example message lowers accuracy to 46.8%.
LatComm compresses LLM context once for reuse across models
On QuALITY the receiver sees 57 times less input and answers an article's questions three times faster.
Safety of Latent Communication in Multi-Agent Systems
Benign latent links between agents raise harmful compliance, and an RL attack lifts the mean score from 27.9 to 76.9.
Draft-KV: Learning Useful Latent Communication Between Language Models
Draft-KV sends a frozen sharer's draft key-value states to a small receiver, lifting MMLU-Redux from 37.45% to 78.04%.
Receiver-Conditioned Latent Communication gives 94% CacheBack
CacheBack filters multi-agent KV caches by receiver need, cutting 75% of state and raising FanOutQA accuracy 14.7 points.
Cache-to-Cache: Direct Semantic Communication Between LLMs (2025)
Research paper proposes Cache-to-Cache, letting LLMs communicate directly via KV-cache semantics instead of tokenized text.