Direct Semantic Communication Between Large Language Models via Vector Translation

TL;DR

This paper introduces a method for direct semantic communication between large language models using learned vector translations, enabling more efficient and meaningful information exchange without token-based messaging.

Contribution

The authors propose a novel vector translation technique for semantic exchange between LLMs, demonstrating its feasibility and stability across different model types.

Findings

Achieved an average cosine alignment of 0.538 with learned mappings.

Injected translated vectors at 30% blending strength to steer model outputs.

Found that general-purpose models have more transferable representations than instruction-tuned models.

Abstract

In multi-agent settings, such as debate, reflection, or tool-calling, large language models (LLMs) pass messages as plain tokens, discarding most latent semantics. This constrains information transfer and adds unnecessary computational overhead. We form a latent bridge via vector translations, which use learned mappings that enable direct semantic exchange between representation spaces. A dual-encoder translator trained between Llama-2-7B and Mistral-7B-Instruct attains an average cosine alignment of 0.538. Injecting the translated vectors at 30 percent blending strength steers the target model's generation without destabilizing logits. Bidirectional evaluation shows a 2.01:1 transfer asymmetry, indicating that general-purpose models yield more transferable representations than instruction-tuned variants. This conservative injection preserves computational stability while demonstrating that cross-model latent communication is feasible, enabling collaborative AI systems that share meaning rather than tokens.