Embeding or other solution to help ChatGPT analyse recent documentation for better code generation

I am trying to use ChatGPT code generation, specifically with AWS’s boto3. However, I tackle many times into issues because it is using an outdated version, and the library is updated constantly and I strongly prefer to use the recent versions.
I thought about downloading the documentations or changelog or bothfrom github and somehow making ChatGPT aware of the latest updates. I read about embedding, but not sure if it’s a good use case as it’s not really “simple search”, and also fine tuning doesn’t seem to be a good fit as I don’t have many examples for each method, but rather one or two that don’t cover all cases.

Any suggestions on how to tackle the issue?


I think crawling the docs, grouping it using gpt-4, filling a vector db maybe in combination with a graphdb to create few shot example prompts might be the way.

That’s a pretty hot thing. Would love to read about your experiments.

1 Like

I encountered this issue while developing a code generation app. I’m planning to find a way to get latest info about library usage to my app.

My idea is to use function calling to let gpt ask for certain library documentation when needed, the function will get the documentation from the internet or a [NEW] project/service that works as index for library documentation.

Currently I put example that includes usage of latest version of a library in my prompt:

But I’m interested in your experiments as well.