Plan Before Search: Search Agents Need Plan
📰 ArXiv cs.AI
Learn how to improve search agents with structured planning, a crucial step before search that enhances retrieval-augmented reasoning
Action Steps
- Define the dependency structure among sub-skills for your search agent
- Implement a structured agentic behavior like Plan for multi-hop retrieval
- Decompose complex tasks into manageable sub-tasks using Plan
- Evaluate the performance of your search agent with and without Plan
- Integrate Plan with reinforcement learning and distillation techniques to enhance capability acquisition
Who Needs to Know This
AI researchers and engineers working on retrieval-augmented reasoning agents can benefit from this approach to improve their models' performance and efficiency
Key Insight
💡 Structured planning is a crucial step before search that can enhance the performance and efficiency of retrieval-augmented reasoning agents
Share This
🔍 Improve search agents with Plan, a structured approach to retrieval-augmented reasoning #AI #SearchAgents
Key Takeaways
Learn how to improve search agents with structured planning, a crucial step before search that enhances retrieval-augmented reasoning
Full Article
Title: Plan Before Search: Search Agents Need Plan
Abstract:
arXiv:2605.28354v1 Announce Type: new Abstract: Training large language models as retrieval-augmented reasoning agents typically combines reinforcement learning with an SFT cold start distilled from a stronger model. However, this paradigm overlooks two fundamental factors: the dependency structure among sub-skills, and the possibility that distillation is not the only route to capability acquisition. We study this through Plan, a structured agentic behavior for multi-hop retrieval that decompos
Abstract:
arXiv:2605.28354v1 Announce Type: new Abstract: Training large language models as retrieval-augmented reasoning agents typically combines reinforcement learning with an SFT cold start distilled from a stronger model. However, this paradigm overlooks two fundamental factors: the dependency structure among sub-skills, and the possibility that distillation is not the only route to capability acquisition. We study this through Plan, a structured agentic behavior for multi-hop retrieval that decompos
DeepCamp AI