Generating responses with new data

Community
, , ,
1

I’m a computer science student working on a project of my own outside of class. Can someone explain to me this article and the significance of it? New embedding models and API updates

Will it help me with my issue?
I am using the openAI API. I have a database full of a lot of data and I want my responses to be based only on the information I am giving it. I previously had an issue, using GPT-4 in my python code that would not follow my instructions.

My instructions are precise. In other words, I have a MySQL database full of the data I want to “teach” the AI system with and give it this information only to generate a response. I am currently concatenating my system instructions var (how to answer) with all of the data.

Each time the user passing in a request I am concatenating system_instructions + system_param (data) each time.

I want to avoid passing all of this data in each time. Furthermore, I read in the API documents that GPT-4 can only read about ~10 pages of data before it ignores information.

Any suggestions on how to handle this? I thought about creating my own LLM. Any feedback would be helpful for a soon to be software engineer.

2

Hi!

are you sure?

:thinking:

a lot of people think that, but then it turns out that often that’s either not the case, or they’re contradictory to boot.

but yes, retrieval augmented generation may help. But what might also help, is to allow the model to generate the SQL query for you. It really depends on your data!