# Integrating GPT3-like bots? **URL:** \ **Category:** Development\ **Created:** [October 22, 2022, 11:45am UTC](https://meta.discourse.org/t/integrating-gpt3-like-bots/242665 "2022-10-22T11:45:49Z")\ **Posts on this page:** 1\ **Showing post:** 8
**Author:** ![simon](https://sea3.discourse-cdn.com/meta/user_avatar/meta.discourse.org/simon/32/339122_2.png) [@simon](https://meta.discourse.org/u/simon)\ **Post date:** [December 4, 2022, 10:18pm UTC](https://meta.discourse.org/t/integrating-gpt3-like-bots/242665/8 "2022-12-04T22:18:53Z")
> [@RGJ](#): > > Do we need a piece of software in between both systems or would a plugin that calls the GPT API suffice? I’ve been mostly off the internet for the past couple of months and came back to see lots of excitement around GPT-3. That got me thinking about how it could be applied to Discourse. I’ve been fooling around with the idea of creating a “staff assistant” bot using OpenAI. I’m testing it with the text-davinci-003 model for now. I _think_ the basic design will be a Discourse post event webhook pointing to a server with the [Python OpenAI library](https://github.com/openai/openai-python) installed on it. The first task I’m working on is getting the bot to return [Data Explorer](https://meta.discourse.org/t/32566?silent=true) queries from plain text descriptions. I haven’t tried fine tuning a model yet, but if information about the Discourse database structure is provided with the request’s prompt, it’s returning surprisingly good results. It may turn out that using OpenAI for this kind of thing will be too expensive, or that the results aren’t accurate enough to use on a production site. I’m just working on this for fun. The other obvious thing to look at is training a customer support bot against existing replies on a Discourse forum. This seems like a more difficult problem (both in terms of cost and accuracy) than just setting up a bot for staff users. --- _[View the full topic](https://meta.discourse.org/t/integrating-gpt3-like-bots/242665)._