mlsanityData to 7 Oct 2026

ALSO: Adversarial Online Strategy Optimization for Social Agents

arXiv 2605.15768 ↗First posted 15 May 2026cs.AI cs.CYcited by 0

On the map: LLM agents and tool use → Reinforcement learning for LLM agents

Abstract (arXiv)

Social simulation provides a compelling testbed for studying social intelligence, where agents interact through multi-turn dialogues under evolving contexts and strategically adapting opponents. Such environments are inherently non-stationary, requiring agents to dynamically adjust their strategies over time. However, most Large Language Model (LLM) based social agents rely on static personas, while existing approaches for enhancing social intelligence, such as offline reinforcement learning or external planners, are ill-suited to these settings, typically assuming stationarity and incurring substantial training overhead. To bridge this gap, we propose ALSO (Adversarial onLine Strategy Optimization), the first framework for online strategy optimization in multi-agent social simulation. ALSO advances social adaptation through two key contributions. (1) ALSO formulates multi-turn interaction as an adversarial bandit problem, where combinations of static personas and dynamic strategy instructions are treated as arms, providing a principled solution to non-stationarity without relying on environmental stability assumptions. (2) To predict rewards and generalize sparse feedback in multi-turn dialogues, ALSO introduces a lightweight neural surrogate to predict rewards from interaction histories, enabling sample-efficient exploration and continuous online adaptation. Experiments on the Sotopia benchmark demonstrate that ALSO consistently outperforms static baselines and existing optimization methods in dynamic environments, validating the effectiveness of adversarial online strategy optimization for building robust social agents.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Reinforcement learning for LLM agents, one of 16 directions in LLM agents and tool use. Builds on 30 of its 53 references that are in the map. First posted on arXiv 15 May 2026.

Builds on

30 of 53 references.

Closest ancestors · 1

Same direction, newest first

  1. 2025SDPO: Segment-Level Direct Preference Optimization for Social Agents3 Jan

Other references · 29

  1. 2024GPT-4o System Cardcited by 4,467
  2. 2023Generative Agents: Interactive Simulacra of Human Behaviorcited by 3,500
  3. 2023A Survey on Large Language Model based Autonomous Agentscited by 2,203
  4. 2023AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversationcited by 1,906
  5. 2022Large Language Models Are Human-Level Prompt Engineerscited by 1,020
  6. 2025OpenAI GPT-5 System Cardcited by 910
  7. 2023Large Language Models as Optimizerscited by 830
  8. 2023Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolutioncited by 521
  9. 2023Understanding the Effects of RLHF on LLM Generalisation and Diversitycited by 386
  10. 2023SOTOPIA: Interactive Evaluation for Social Intelligence in Language Agentscited by 350
  11. 2024Optimizing Instructions and Demonstrations for Multi-Stage Language Model Programscited by 270
  12. 2022Neural Theory-of-Mind? On the Limits of Social Intelligence in Large LMscited by 220
17 more
  1. 2023Theory of Mind for Multi-Agent Collaboration via Large Language Modelscited by 129
  2. 2024Systematic Biases in LLM Simulations of Debatescited by 98
  3. 2024The Butterfly Effect of Altering Prompts: How Small Changes and Jailbreaks Affect Large Language Model Performancecited by 97
  4. 2024SOTOPIA-π: Interactive Learning of Socially Intelligent Language Agentscited by 89
  5. 2023AI model GPT-3 (dis)informs us better than humanscited by 75
  6. 2023Testing the Reliability of ChatGPT for Text Annotation and Classification: A Cautionary Remarkcited by 54
  7. 2024SocialBench: Sociality Evaluation of Role-Playing Conversational Agentscited by 48
  8. 2024Prompt Optimization with EASE? Efficient Ordering-aware Automated Selection of Exemplarscited by 37
  9. 2024AgentSense: Benchmarking Social Intelligence of Language Agents through Interactive Scenarioscited by 37
  10. 2024Prompt Optimization with Human Feedbackcited by 35
  11. 2023Use Your INSTINCT: INSTruction optimization for LLMs usIng Neural bandits Coupled with Transformerscited by 29
  12. 2025Sotopia-RL: Reward Design for Social Intelligencecited by 20
  13. 2026The Need for a Socially-Grounded Persona Framework for User Simulationcited by 14
  14. 2024Dialogue Action Tokens: Steering Language Models in Goal-Directed Dialogue with a Multi-Turn Plannercited by 12
  15. 2025EPO: Explicit Policy Optimization for Strategic Reasoning in LLMs via Reinforcement Learningcited by 8
  16. 2025Meta-Prompt Optimization for LLM-Based Sequential Decision Makingcited by 7
  17. 2025Infusing Theory of Mind into Socially Intelligent LLM Agentscited by 7

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026One Model, All Roles: Multi-Turn, Multi-Agent Self-Play Reinforcement Learning for Conversational Social Intelligence
  2. 2026AI Agents Alone Are Not (Yet) Sufficient for Social Simulation
  3. 2026PB-GRPO: Learning Socially Adaptive LLM Agents from Persona-Driven Simulation with Preference-Batched GRPO
  4. 2026Playing social deduction games with reinforcement fine-tuned large language models
  5. 2025Enhancing Language Agent Strategic Reasoning through Self-Play in Adversarial Games
  6. 2026SocialRL: Refining LLMs' Social Intelligence through Multi-turn Reinforcement Learning and Reward Design
  7. 2025Learning Robust Social Strategies with Large Language Models
  8. 2026SODE: Analyzing Social Dynamics in LLM Agents