@adrienne

Just pushed an update that will improve how Warpee answers questions about specific episodes. The root cause of the problem was a filtering function I built whose job was to try to shrink our long list of episodes before sending it to the LLM. It was too aggressive and filtered out too many episodes so the LLM didn't have the correct context to answer. When I first built Warpee, I was using models with smaller context windows so had to be careful about not sending too much data into the prompt. I'm using GPT-5 now which allows much larger context window and is cheaper, so I just removed that filtering step and now Warpee gets the full list of episodes to help him answer questions. Feel free to try it out by asking him questions like "what was our latest episode" or "give me some highlights from our last episode"

3 replies

0 recast

2 reactions