Adding custom info to GPT without retrieval

I want to incorporate knowledge about a specific domain into a GPT-4 based application.

I’m wondering if there are any other options than prompting / retrieval augmented generation, to make the model aware of specific knowledge.

It’s possible to fine-tune GPT3.5 now, and fine-tuning on GPT4 just started but is prohibitively expensive.

Is there any other way to teach GPT4 new things than injecting custom information into a prompt (via retrieval)?

With the API brain being a closed model (ie cant access internet), the only ways to provide it with information is via the RAG or via providing it information via the prompt

1 Like