OpenAI made quite a splash when they announced a slew of products at Dev Day. In particular, two announcements have sent people crazy: the Assistants API and GPTs. Many now claim that users have no reason to use third-party Vector DBs / RAG tools for querying their data when OpenAI now provides RAG natively. While GPTs are less of a threat to RAG startups because they are solely consumer-oriented, the Assistant API is since it could act as a drop-in replacement for existing RAG pipelines within businesses. However, there's one key question everyone is failing to ask: Is OpenAI's system even good? Short answer: No. As you'll soon see, while OpenAI's promises are big, the execution leaves much to be desired.
As with everything in programming, the first step to getting started building something is to RTFM (read the f*cking manual). Unfortunately, OpenAI’s instructions leave a lot to be desired. For instance, they describe how to create an assistant but don’t give you full instructions on retrieving messages. Instead, they only give you instructions on retrieving this giant and complex object with messages hidden inside it.
Sadly, this was not in the docs, and I had to figure it out myself by manually inspecting the messages object in the VsCode debugger.