发表机构
Nanyang Technological University（南洋理工大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究提出REFLEX架构，利用快速类型化决策层Jev在低置信度时调用强LLM，在100任务基准上以95%成功率减少72.7%强模型调用，并揭示了其适用边界。
AI 中文摘要
LLM智能体通常使用生成模型进行有界决策，这引发了一个问题：在何种情况下，这些决策可以在不降低任务成功率的前提下被更高效地处理。我们研究了REFLEX，一种智能体架构，它使用Jev作为快速的、类型化的决策层，并在置信度较低或需要生成时调用强大的LLM。在一个固定的100任务基准上，REFLEX实现了95%的成功率，同时相比仅使用强模型的智能体，强模型调用次数减少了72.7%，且这一减少在三种回退家族中持续存在。受控干预实验表明，可靠性取决于动作集大小以及接近授权边界的近似有效备选项。外部BFCL和τ风格评估显示，当常规路由已经高度准确时，与廉价的生成级联相比，其优势有限。这些发现明确了Jev选择性控制何时能减少计算量，以及其优势在何处受限。
英文摘要
LLM agents often use generative models for bounded decisions, raising the question of when these decisions can be handled more efficiently without reducing task success. We study REFLEX, an agent architecture that uses Jev as a fast, typed decision layer and calls a strong LLM when confidence is low, or generation is required. On a frozen 100-task benchmark, REFLEX achieves 95% success with 72.7% fewer strong-model calls than a strong-only agent, with reductions persisting across three fallback families. Controlled interventions show that reliability depends on action-set size and near-valid alternatives near authorization boundaries. External BFCL and $τ$-style evaluations reveal limited advantages over a cheap generative cascade when ordinary routing is already highly accurate. These findings identify when selective control with Jev can reduce computation and where its benefits are limited.
