Hello! I’ve been exploring the newly launched Assistants features and I’m curious if the context provided for each Assistant is similar to fine-tuned training for specific entities.
Can an Assistant also incorporate fine-tuning files or training, or are they mutually exclusive? Could someone clarify this for me, please?
Following this thread! Would be interesting to get more info on how assistants (with context and files) compare to a finetuned model. Seems like assistants might be a better way to go since they allow for multiple tools to use!
+1
I just setup a support agent assistant to read my knowledge base. Its really close to being accurate - and I feel like if there was a tuning option it could get there quicker.
Right now, I’d have to rewrite many articles to make them more AI friendly at the expense of being less clear to humans
+1
This feature enables the setup of an assistant with file retrieval AND a finely tuned model to generate the answer.
Often, the documentation from OpenAI itself explains that fine-tuning and data-retrieval are complementary approaches.
However, I still can’t see how I can leverage both of them in the same Assistant.
I’ll try to do the code now i don think it could hurt to much in the pockets if the systems are complementary.
The solution i would implement is to have three open ai assistants, one make function calls to to:
To create assistants - https://platform.openai.com/docs/assistants/overview
I’ll go read up on fine-tuning now
https://platform.openai.com/docs/api-reference/fine-tuning/create
P.s
Im tapping out early. Seems to be for commercial use. It says we need samples of a minimum of 50-100 high quality examples to train the model in batches.
It suggests focusing on getting good at prompt completions, which i don’t have yet, that will translate into the example data for training a model.
I figure when i get 200-300 specific prompts I’ll give it a go