Hey everyone!! I heard around the term "DIY RAG" can someone give me light on it and where it would be found in graphic?

@Jorge Arone RAG = Retrieval Augmented Generation. It is a method to enhance an LLM responses for specific tasks. For instance, if you're a company with a knowledge exchange, you might tokenize the entire knowledge catalog in a vector database and use that to inform a foundation model. This is not training a new model as fine-tuning requires, merely focusing the responses from a foundation model to be informed and anchored to the information in the knowledge catalog. It is significantly faster and cheaper (for now) to do RAG than fine-tuning. There are other methods as well, such as Forward Looking Active REtrieval augmented generation (FLARE). Here's one of many deeper conversations. So, DIY RAG probably means someone has made it pretty easy to do RAG. There's actually quite a few startups doing this stuff today. It's not terribly complex and it probably one of the easier and more straightforward ways to "customize" or constrain an LLM to be informed by domain-specific data. As this graphic shows, there are MANY methods to choose from (many in the graphic are new to me too). So, just like traditional AI/ML where the algorithms used to create the model mattered, the technique and tool you use to customize the output of a foundation LLM model matters too. Good luck!