From Words to Widgets for Controllable LLM Generation

arXiv cs.CL · 18 hours ago

Researchers proposed Malleable Prompting, an interactive technique that converts natural language preference expressions in LLM prompts into GUI widgets like sliders and dropdowns for more precise control over generation output. A user study demonstrated that the approach achieved target preferences more precisely than natural language prompting alone, with participants perceiving it as more controllable and transparent. This method enables users to directly configure generation parameters and visualize how each control influences the output across iterations.