GPT3 & Live Data | OpenAI | Page 1

wheat musk Dec 28, 2022, 10:42 PM

#

GPT-3 does not have the ability to access live data.

#

GPT3 & Live Data

drifting heron Dec 28, 2022, 10:56 PM

#

@wheat musk , thanks, so there would need to be something in between, like for example process a request to generate some sort of formatted query then feed that to a secondary system?

wheat musk Dec 28, 2022, 10:57 PM

#

hmmm. It really depends on what you're trying to do. It's not totally clear based on what you've written. Can you give one example of an end-to-end flow?

drifting heron Dec 28, 2022, 10:58 PM

#

sure, let's say I want to ask about the state of a number of live tracked sensors that all generate real time data

#

but I don't want the person asking to be writing a query to pull data from a database or anything like that

#

"I want to know which sensors are operating outside the norm or are experiencing unusual activity"

#

then we have a database with live sensor data, that has this information, that we want to pull

#

as you mentioned, there's no way for GPT3 to know anything about this data, but it could generate a query to pull the data and then we could output this separately based on the query?

wheat musk Dec 28, 2022, 11:02 PM

#

interesting, good use case....
GPT-3 can only answer questions about what it has knowledge of. There isn't a way to feed in live data.

It sounds like what you really want is to be able to query a database without the user having to use a database query language. Correct?

drifting heron Dec 28, 2022, 11:03 PM

#

something along those lines. And I understand reading the docs that this would be more of a fine-tuning type of scenario, but essentially it would be a use-case specific query with a specific format to be interpreted by a secondary system given the limitation of access to that live data

wheat musk Dec 28, 2022, 11:05 PM

#

Yes I think you are right. I'm imagining you could do this, though it would be a bit ham fisted and potentially prone to error (unless you fine tune a model):

User types in natural language what they want to know.
You design a backend that will take that and merge in some information about the database (eg. column names, row names)
The back end sends this to GPT requesting an appropriate query for the database (eg. SQL)
GPT returns the result to your back end
Your back end submits the query
Your back end returns the answer to the database query for the user

drifting heron Dec 28, 2022, 11:06 PM

#

I appreciate you taking the time to answer this.

wheat musk Dec 28, 2022, 11:07 PM

#

You're welcome. There may be better ways to do this! There are many ways to skin a cat... But that is what comes to mind for me.

drifting heron Dec 28, 2022, 11:08 PM

#

I'll keep digging through the docs and see what comes up, Thanks again, really appreciated

wheat musk Dec 28, 2022, 11:08 PM

#

If you end up designing this or come up with a better system please share here! I'd love to see how it works or if there is a better way to do this 🙂

#GPT3 & Live Data