Verifying Function Selection and Argument Values in ELMs

In this article, the author examines a critical aspect of integrating tools into chat-bots based on Large Language Models (ELMs). The focus is on two scenarios: when the model must execute a function call and when none of the available tools are suitable for the task. The author provides a comparative analysis of two implementation approaches. The first method involves the model generating the function call independently, formulating arguments on the fly. The second method involves selecting a function and argument values from a pre-defined list of options. The article helps developers understand which approach provides greater reliability and accuracy when working with external APIs, minimizing errors in interpreting user intent. This guide is useful for professionals involved in building agentic systems and integrating LLMs into application programming interfaces.