Llm agents balance internal knowledge with external information
Look Before You Leap: Thermodynamic Arbitration of Parametric and Non-Parametric Knowledge in LLM Agents via Self-Regulating Memory Architectures
Artificial IntelligenceComputation and Language
Summary
Many AI systems that use large language models rely heavily on checking outside information for every response, which can be slow and confusing. The authors propose a new design that lets these AI agents first think about how certain they are of their own knowledge before looking for external help. This approach is inspired by how living beings 'look before they leap,' meaning they check carefully before acting. Their system, called MARTA, helps AI be more thoughtful and efficient by using both what they already know and information they retrieve only when necessary.
What this means in practice
- •For chatbot developers: Design chatbots that decide when to use their own knowledge versus external data to improve response efficiency and relevance.
- •For digital assistant engineers: Build digital assistants that reduce unnecessary data retrieval by assessing their certainty before consulting external sources.
Authors
Akash Das, Ishan Roy
Abstract
The architecture of modern LLMs consists of a profound cognitive polarization. LLMs possess implicit intuition encoded in their parameters, yet rely on a disconnected, explicit mechanism to access the outside world. Agentic frameworks have not bridged this gap; instead, models are often compelled into pathological "induced amnesia." Under the prevailing "Retrieve-Always" paradigm, agents must distrust their internal knowledge, making every user interaction a "tabula rasa" event that must be checked externally. This creates reflexive dependence that can be thermodynamically wasteful, cognitively fragile, and susceptible to irrelevant context. We propose a return to first principles, operationalizing the biological maxim "Look Before You Leap." We introduce MARTA (Metacognitive Adaptive Retrieval and Thought Architecture), a neuro-symbolic framework that bridges parametric and non-parametric knowledge. Rather than treating retrieval as mandatory, MARTA models it as a cost, taking the leap only when perceived internal inadequacy warrants external information. By allowing the agent to gauge the entropy of its own thoughts before acting, MARTA enables deliberative retrieval and uncertainty-aware decision making. Our approach suggests that giving agents the capacity for introspection can restore a more efficient balance between internal knowledge and external information.