So I’ve written a plugin for the Discourse forum platform (the same that serves this community) to summarise Topics (like this one) using an Open AI model via the official API and one of the popular Ruby API libraries (thanks guys!)
By default it uses text-davinci-003
, but you can choose the model.
This is very experimental but seems stable after a couple of days trialling it.
The results are surprisingly good so far, and occasionally sublime!
However, whilst it often returns relevant, syntactically correct and convincing summaries, it is prone to factual inaccuracies that can be highly misleading and impair its usefulness. So much potential here though!
Note due to API token limits, it will only cover up to about ~40 Posts (less for less capable models), but that’s pretty good already imho and definitely usable. There is a minimum threshold setting to ensure a summary is not sought until you have a decent enough amount of material.
There’s a downvoting button & mechanic to allow the community to force the system to make another request and various quality of life settings.
Of course if you want to try it out you will need to add this plugin to a Discourse forum instance, although these can be run locally.
If you have any suggestions for how to improve this I’m all ears!