{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/d554c62f8cbf446580712a9ff07b18de\" frameborder=\"0\" width=\"1960\" height=\"1470\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1470,"width":1960,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1470,"thumbnail_width":1960,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/d554c62f8cbf446580712a9ff07b18de-00001.gif","duration":255.53333333333325,"title":"Few-shot prompting - Part 2: Creating a dataset that's used by the LLM","description":"Hi! In this video, I'll show you how to prepare a data set for Relevance AI to use in few-shot prompt examples. I'll be taking some replies and responses that we've given to LinkedIn messages, but you can collect data in any way that you find convenient. The process of creating this data set is pretty much the same, regardless of what data you have. We'll start in a spreadsheet, create header columns for every field we want to use, and fill it up with data. Once we're ready, we'll save and download it as a CSV, and then upload it to the Relevantize Dashboard. We'll vectorize the text so that we can do semantic search on our responses, and then use it in the similarity search step in a chain. It might take a few minutes to import and vectorize the data, but once it's done, we can use it to match up the nearest response to a set of responses we have in the past. No action is requested from you, but I hope you find this helpful!"}