Latent Agents: A Post-Training Procedure for Internalized Multi-Agent Debate

• URL: http://arxiv.org/abs/2604.24881v1
• Date: Mon, 27 Apr 2026 18:06:03 GMT
• Title: Latent Agents: A Post-Training Procedure for Internalized Multi-Agent Debate
• Abstract summary: Multi-agent debate has been shown to improve reasoning in large language models (LLMs)<n>We develop a framework that distills multi-agent debate into a single LLM through a two-stage fine-tuning pipeline.<n>Our internalized models match or exceed explicit multi-agent debate performance using up to 93% fewer tokens.
• Score: 16.68699018207298
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Multi-agent debate has been shown to improve reasoning in large language models (LLMs). However, it is compute-intensive, requiring generation of long transcripts before answering questions. To address this inefficiency, we develop a framework that distills multi-agent debate into a single LLM through a two-stage fine-tuning pipeline combining debate structure learning with internalization via dynamic reward scheduling and length clipping. Across multiple models and benchmarks, our internalized models match or exceed explicit multi-agent debate performance using up to 93% fewer tokens. We then investigate the mechanistic basis of this capability through activation steering, finding that internalization creates agent-specific subspaces: interpretable directions in activation space corresponding to different agent perspectives. We further demonstrate a practical application: by instilling malicious agents into the LLM through internalized debate, then applying negative steering to suppress them, we show that distillation makes harmful behaviors easier to localize and control with smaller reductions in general performance compared to steering base models. Our findings offer a new perspective for understanding multi-agent capabilities in distilled models and provide practical guidelines for controlling internalized reasoning behaviors. Code available at https://github.com/johnsk95/latent_agents

Related papers

• Walk the Talk: Bridging the Reasoning-Action Gap for Thinking with Images via Multimodal Agentic Policy Optimization [89.68681087743876]
  Multimodal Large Language Models (MLLMs) have incentivized models to think with images'' by actively invoking visual tools during multi-turn reasoning.<n>The common Reinforcement Learning practice of relying on outcome-based rewards ignores the fact that textual plausibility often masks executive failure.<n>We introduce Multimodal Agentic Policy Optimization (MAPO), bridging the gap between textual reasoning and visual actions generated by models within their Multimodal Chain-of-Thought.
  arXiv  Detail & Related papers  (2026-04-08T07:48:07Z)
• DLLM Agent: See Farther, Run Faster [94.74432470237817]
  Diffusion large language models (DLLMs) have emerged as an alternative to autoregressive (AR) decoding with appealing efficiency and modeling properties.<n>We study this in a controlled setting by instantiatingDLLM and AR backbones within the same agent workflow.<n>We find thatDLLM Agents are on average over 30% faster end to end than AR agents, with some cases exceeding 8x speedup.
  arXiv  Detail & Related papers  (2026-02-07T09:01:18Z)
• AgentArk: Distilling Multi-Agent Intelligence into a Single LLM Agent [57.10083973844841]
  AgentArk is a novel framework to distill multi-agent dynamics into the weights of a single model.<n>We investigate three hierarchical distillation strategies across various models, tasks, scaling, and scenarios.<n>By shifting the burden of computation from inference to training, the distilled models preserve the efficiency of one agent while exhibiting strong reasoning and self-correction performance of multiple agents.
  arXiv  Detail & Related papers  (2026-02-03T19:18:28Z)
• MADIAVE: Multi-Agent Debate for Implicit Attribute Value Extraction [52.89860691282002]
  Implicit Attribute Value Extraction (AVE) is essential for accurately representing products in e-commerce.<n>Despite advances in multimodal large language models (MLLMs), implicit AVE remains challenging due to the complexity of multidimensional data.<n>We introduce textscmodelname, a multi-agent debate framework that employs multiple MLLM agents to iteratively refine inferences.
  arXiv  Detail & Related papers  (2025-10-07T06:27:42Z)
• Tree of Agents: Improving Long-Context Capabilities of Large Language Models through Multi-Perspective Reasoning [11.045096250408067]
  Tree of Agents (TOA) is a multi-agent reasoning framework that segments the input into chunks processed by independent agents.<n>TOA enables agents to probe different reasoning orders for multi-perspective understanding.<n>To improve processing efficiency, we incorporate prefix-hash caching and adaptive pruning strategies.
  arXiv  Detail & Related papers  (2025-09-08T08:34:02Z)
• Distilling LLM Agent into Small Models with Retrieval and Code Tools [65.73762766854192]
  Agent Distillation is a framework for transferring reasoning capability and task-solving behavior from large language models into small language models.<n>Our results show that sLMs as small as 0.5B, 1.5B, 3B parameters can achieve performance competitive with next-tier larger 1.5B, 3B, 7B models.
  arXiv  Detail & Related papers  (2025-05-23T08:20:15Z)
• Enhancing Video-LLM Reasoning via Agent-of-Thoughts Distillation [32.930999188946345]
  This paper tackles the problem of video question answering (VideoQA)<n>Large video-language models perform well on benchmarks, but they often lack explainability and spatial-temporal grounding.<n>We propose Agent-of-Thoughts Distillation (AoTD), a method that enhances models by incorporating automatically generated Chain-of-Thoughts (CoTs) into the instruction-tuning process.
  arXiv  Detail & Related papers  (2024-12-02T16:37:50Z)
• MADiff: Offline Multi-agent Learning with Diffusion Models [79.18130544233794]
  MADiff is a diffusion-based multi-agent learning framework.<n>It works as both a decentralized policy and a centralized controller.<n>Our experiments demonstrate that MADiff outperforms baseline algorithms across various multi-agent learning tasks.
  arXiv  Detail & Related papers  (2023-05-27T02:14:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.