There's something interesting going on where the LLM API returns 5 replies that have a "blendedScore", then the frontend chooses one and displays it based on th

There's something interesting going on where the LLM API returns 5 replies that have a "blendedScore", then the frontend chooses one and displays it based on those scores It might be doing N generations with self-scoring, then averaging the scores after the fact