Towards Communication-Efficient Social Intelligence in Language Agents
What it is
This paper proposes Teacher-Assisted Communication Training (TACT), a method to improve the social intelligence of language agents by balancing goal attainment with communication efficiency. It uses a teacher model to revise student actions and distill feedback, enabling agents to convey necessary information without unnecessary verbosity.
The problem it solves
Current language agents often fail to balance the need to convey sufficient information for coordination with the need to minimize communication cost, leading to inefficient or failed social interactions.
What you could use it for
- Designing conversational agents for customer service that resolve conflicts quickly
- Building negotiation agents that minimize interaction turns
- Improving multi-agent coordination systems where communication bandwidth is limited
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.