Hi, I’ve built a Pickaxe to do simple image recognition. It’s consistently messing up one of the images despite the information being included in the knowledge base. How do I tell the model it’s wrong and what the correct response should be?
There are multiple ways to do this. We do not have an automatic feedback system, though we have the infrastructure in place to build it eventually (i.e. we have front-end UI to collect upvote/downvote).
What I do with the Pickaxe doorman is build out and a continuously add to a spreadsheet with good example answers. I’ll enter the question and then the right answer. I briefly go over it in this video about building the Pickaxe doorman.