Główne pojęcia
A novel human-LLM interaction framework, Low-code LLM, incorporates simple low-code visual programming interactions to achieve more controllable and stable responses from LLMs for complex tasks.
Streszczenie
The paper introduces a novel human-LLM interaction framework called Low-code LLM. It aims to improve the control and efficiency of utilizing large language models (LLMs) for complex tasks.
The key components are:
Planning LLM: Generates a structured planning workflow for complex tasks, which can be edited and confirmed by users through low-code visual programming operations.
Executing LLM: Generates responses by following the user-confirmed workflow, enabling more controllable and satisfactory results.
The low-code interaction allows users to easily understand and modify the logic and workflow underlying the LLMs' execution, bridging the gap between humans and LLMs. This framework offers three main advantages:
User-friendly Interaction: The visible workflow provides users with a clear understanding of how LLMs execute tasks, enabling easy editing through a graphical user interface.
Controllable Generation: Complex tasks are decomposed into structured workflows, allowing users to control the LLMs' execution through low-code operations.
Wide Applicability: The framework can be applied to various complex tasks across domains, especially where human intelligence or preferences are critical.
The paper demonstrates the benefits of Low-code LLM through four pilot cases: essay writing, object-oriented programming, virtual hotel service, and resume helper. These experiments show how Low-code LLM enables more controllable and satisfactory results compared to conventional prompt engineering.