AI Engineering
My Weird Prompts is made by an AI pipeline, and we share how it’s built: the pipeline itself, what we changed and why, and the evaluations we run to choose the models that write the show.
- New Evals The show’s own evaluations, open-sourced. ScriptBench, the script-writer benchmark, is the first.
- ScriptBench Live leaderboard, every model that has written the show, episodes by model and the model chronicle.
- Technical White Paper The production pipeline end to end: stages, models, costs, safeguards and live stats.
- Pipeline Changelog What changed in the pipeline, and when.
- How It Works The short version: from a voice memo to a finished episode.
- Data Portal Open datasets: episodes, transcripts and the model chronicle.
- Public MCP Server Query the catalogue from your own agent.
- eval repo on GitHub ↗ ScriptBench’s harness, results and working paper.