![[2602.19458] ComplLLM: Fine-tuning LLMs to Discover Complementary Signals for Decision-making](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19458] ComplLLM: Fine-tuning LLMs to Discover Complementary Signals for Decision-making
Summary
The paper presents ComplLLM, a framework for fine-tuning large language models (LLMs) to enhance decision-making by utilizing complementary signals from multiple agents.
Why It Matters
As decision-making increasingly relies on AI, understanding how to leverage complementary information from various agents can significantly improve outcomes. This research introduces a novel approach that could enhance the effectiveness of AI systems in complex decision environments, making it relevant for both AI developers and decision-makers.
Key Takeaways
- ComplLLM fine-tunes LLMs using complementary information as rewards.
- The framework demonstrates improved decision-making in multi-agent scenarios.
- Validation on synthetic and real-world tasks shows the framework's effectiveness.
- The approach provides plausible explanations for complementary signals.
- This research contributes to the field of decision theory in AI.
Computer Science > Artificial Intelligence arXiv:2602.19458 (cs) [Submitted on 23 Feb 2026] Title:ComplLLM: Fine-tuning LLMs to Discover Complementary Signals for Decision-making Authors:Ziyang Guo, Yifan Wu, Jason Hartline, Kenneth Holstein, Jessica Hullman View a PDF of the paper titled ComplLLM: Fine-tuning LLMs to Discover Complementary Signals for Decision-making, by Ziyang Guo and 4 other authors View PDF HTML (experimental) Abstract:Multi-agent decision pipelines can outperform single agent workflows when complementarity holds, i.e., different agents bring unique information to the table to inform a final decision. We propose ComplLLM, a post-training framework based on decision theory that fine-tunes a decision-assistant LLM using complementary information as reward to output signals that complement existing agent decisions. We validate ComplLLM on synthetic and real-world tasks involving domain experts, demonstrating how the approach recovers known complementary information and produces plausible explanations of complementary signals to support downstream decision-makers. Subjects: Artificial Intelligence (cs.AI); Human-Computer Interaction (cs.HC) Cite as: arXiv:2602.19458 [cs.AI] (or arXiv:2602.19458v1 [cs.AI] for this version) https://doi.org/10.48550/arXiv.2602.19458 Focus to learn more arXiv-issued DOI via DataCite (pending registration) Submission history From: Ziyang Guo [view email] [v1] Mon, 23 Feb 2026 03:01:52 UTC (1,017 KB) Full-text links: Access ...