Research5h ago
GROVE Visualizes Language Model Output Distributions
A study of 13 researchers led to GROVE, an interactive tool that maps multiple LLM generations as overlapping paths in a text graph. It exposes hidden modes and edge cases that single-output interactions typically hide. This allows developers to stop over-generalizing from anecdotal samples. Practitioners can now visualize stochasticity to refine prompts for open-ended tasks.