智能体
arxiv.org ·
合作覆盖回路如何抑制大语言模型中的纳什博弈
arXiv 论文研究 LLM 在囚徒困境中的合作行为机制。在直接提示下，Llama-3-70B、Qwen2.5-32B 与 Qwen2.5-72B 锁定完全合作（与纳什距离最大、重复间零方差），Llama-3-8B 接近纳什。logit-lens 分析显示存在分布式“合作覆盖”：约四分之三网络深度的中间读出偏向纳什动作，末层才转向合作，且末层修正幅度与思维链行为在规模和两种架构上秩匹配。在 8B 模型中，该覆盖是残差流中单一可因果操控方向，钳制其某一层某一位置的分量可使选择严格单调变化（Spearman rho=1.000）且生成流畅。该回路是词汇性的：去除名称与收益缩放后仍存在，但把 Cooperate/Defect 换成中性标签后失效；在 48 个中性表面收益随机博弈中，无模型在任何困境上锁定合作或展现一般均衡能力。混合模型群体中，单个纳什博弈智能体会传染性地瓦解合作。作者认为抑制纳什博弈的是词触发回路而非能力缺失，且可测量、可界定、可控制。
AI推荐理由 arXiv 论文（v3，2026-09-18 修订），摘要给出具体实验设置与量化结果：Llama-3-70B、Qwen2.5-32B/72B 在直接提示的囚徒困境中锁定完全合作且零方差，Llama-3-8B 接近纳什；logit-lens 发现分布式“合作覆盖”回路，约四分之三网络深度偏向纳什动作后末层转向合作；8B 中该覆盖是残差流中单一可因果操控方向，钳制后选择严格单调（Spearman rho=1.000）；回路是词汇性的，替换中性标签后失效，48 个中性表面收益随机博弈中无模型锁定合作。新信息在于把 LLM 合作行为归因于词触发回路而非能力缺失，并给出可测量、可操控的机制证据；局限是仅覆盖少数模型与特定博弈框架，摘要未提供统计细节与外部复现。
78
正文
Computer Science > Computer Science and Game Theory
arXiv:2604.27167 (cs)
[Submitted on 29 Apr 2026 (v1), last revised 18 Sep 2026 (this version, v3)]
Title:How a Cooperative-Override Circuit Suppresses Nash Play in Large Language Models
Authors:Paraskevas V. Lekeas, Giorgos Stamatopoulos
View a PDF of the paper titled How a Cooperative-Override Circuit Suppresses Nash Play in Large Language Models, by Paraskevas V. Lekeas and Giorgos Stamatopoulos
View PDF
HTML (experimental)
Abstract:On the named Prisoner's Dilemma under direct prompting, three larger instruction-tuned models, Llama-3-70B, Qwen2.5-32B, and Qwen2.5-72B, lock at full cooperation, the metric's maximum distance from Nash with zero variance across replicates, while Llama-3-8B plays near-Nash. Opening the models, a logit-lens analysis finds a distributed cooperative override. Intermediate readouts lean toward the Nash action through roughly three quarters of network depth before a late surge toward cooperation, and the final layer settles the contest. The size of that final correction, not the surge, rank-matches chain-of-thought behavior across scale and two architectures. In the 8B the override is a single causally controllable direction in the residual stream; steering it dials the decision, and clamping its component at one position of one layer moves the choice strictly monotonically, Spearman rho = 1.000, with generation fluent. The circuit is lexical. It survives name removal and payoff rescaling but disengages when Cooperate and Defect are replaced with neutral labels, and on 48 payoff-random games with neutral surfaces no model locks cooperative on any dilemma or shows general equilibrium competence. In mixed-model populations a single Nash-playing agent collapses cooperation contagiously. What suppresses Nash play in large language models is a word-triggered circuit rather than missing competence, and it can be measured, bounded, and controlled.
Comments:
Subjects:
Computer Science and Game Theory (cs.GT); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
MSC classes:
91A10, 68T07
ACM classes:
I.2.6; I.2.11; J.4
Cite as:
arXiv:2604.27167 [cs.GT]
(or
arXiv:2604.27167v3 [cs.GT] for this version)
https://doi.org/10.48550/arXiv.2604.27167
Focus to learn more
arXiv-issued DOI via DataCite
Submission history
From: Paraskevas Lekeas [view email]
[v1]
Wed, 29 Apr 2026 20:16:54 UTC (199 KB)
[v2]
Mon, 4 May 2026 01:02:42 UTC (217 KB)
[v3]
Fri, 18 Sep 2026 15:00:08 UTC (109 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled How a Cooperative-Override Circuit Suppresses Nash Play in Large Language Models, by Paraskevas V. Lekeas and Giorgos Stamatopoulos
View PDF
HTML (experimental)
TeX Source
view license
Current browse context:
cs.GT
new
|
recent
| 2026-04
Change to browse by:
cs
cs.AI
cs.LG
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
export BibTeX citation
Loading...
BibTeX formatted citation
×
loading...
Data provided by:
Bookmark
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Code, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos
Demos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers
Recommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
Author
Venue
Institution
Topic
About arXivLabs
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Which authors of this paper are endorsers? |
Disable MathJax (What is MathJax?)