CoordiLang: Assessing Multi-Agent Coordination Skills in Large Language Models
Bibliographic record
Abstract
The advanced reasoning and inferential capabilities exhibited by Large Language Models (LLMs) position them as viable candidates for orchestrating coordination among multiple agents. This paper presents \textit{CoordiLang}, a novel benchmark designed to rigorously evaluate the coordination prowess of LLMs within the framework of Pure Coordination Games, where agents must collaborate without conflicting interests to maximize collective gains. \textit{CoordiLang} encompasses two primary evaluation facets: (1) \textbf{Agentic Coordination}, wherein LLMs assume proactive roles in facilitating cooperation across four distinct pure coordination scenarios; and (2) \textbf{Coordination Question Answering (QA)}, involving 200 meticulously crafted multiple-choice queries derived from the aforementioned games to assess three critical reasoning dimensions: Environmental Understanding, Theory of Mind (ToM) Reasoning, and Collaborative Planning. Additionally, we introduce the \textit{Coordination Cognitive Framework (CCF)}, a modular architecture enabling seamless integration of various LLMs as interchangeable components within coordination tasks. Empirical results demonstrate that LLMs, particularly those augmented with the latest iterations like GPT-4-X, achieve performance on par with state-of-the-art reinforcement learning (RL) agents in environments necessitating intuitive, environment-based actions. Notably, zero-shot coordination assessments reveal that LLMs exhibit enhanced adaptability to novel partners compared to traditional RL methodologies. However, significant gaps remain in their ToM reasoning and collaborative planning capabilities, highlighting avenues for future improvement. Our comprehensive analysis underscores the pivotal role of environmental comprehension and partner intention inference in effective multi-agent collaboration.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.006 | 0.034 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.004 |
| Open science | 0.003 | 0.003 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.004 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".