Show HN: SlopGrid, a No-Code Tool for One-Shot AI Benchmarks

Summary

SlopGrid is a no-code app for arranging one-shot benchmarks for large language models, with a strong focus on mobile use. It can compare model outputs in text and images, as well as SVG files and complete static websites. The creator says the website format makes the benchmarks especially interesting because it allows models to be assessed on more than conventional text responses. SlopGrid also lets users alter chat history in the chat-completion API, a feature intended for model red-teaming and for examining how models respond in unusual conversational contexts. The project was inspired by Simon Willison’s benchmark writeups. The creator says using it has improved their intuition about the strengths and weaknesses of current leading models, and has published examples for others to inspect or recreate.