隆重推出 Discourse AI

[Falco]

We are happy to announce Discourse AI, a new plugin and our one-stop solution for integrating Artificial Intelligence and Discourse, enabling both new features and enhancing existing ones. With this first release, we are shipping 7 different Discourse AI modules to help community managers, members, and moderators with various tasks from sentiment analysis to automated proofreading and suggested edits. Read along and find out more details about each of these features as well as what is coming up next on our roadmap!

---

This is a companion discussion topic for the original entry at https://blog.discourse.org/2023/04/introducing-discourse-ai/

[justin (Justin DiRose)]

This is an impressive body of work, @Falco and team. Really excited to see how this all works in practice and its impact on community management overall.

[ozkn (Mathx)]

Thanks for the plugin. I’m looking forward to the feature to extract text from images.

[jordan-violet]

These are the kind of updates that feel like opening a new Christmas present.

We (at the present time of this writing) do not have a dedicated manager for our community, and tools like this enable us to continue to scale without a dedicated role.

Not to mention the features like composer helper that just elevate the user experience.

[Helga_Razinkova (Helga)]

Wow, it sounds awesome! Looking forward to having the AI feature on Business plan

[rosiesherry]

I’m excited for this and adding it in as soon as I can.

Are there any plans to use AI to detect AI post/comment spam? It’s a huge problem (in general) in communities.

[Jagster (Jakke Lehtonen)]

Nice! But… why are nice things everytime so god damn expensive

[Tris20 (Tristan)]

@Falco

Extraction of text from images

I like this one

[Canapin (Coin-coin le Canapin)]

Well, if you want to generate text or have answers, there’s the cheaper old-school version

[sam (Sam Saffron)]

Hi Rosie,

Yes we are planning on exploring this area. The tricky thing is that we only have a small number of examples to feed into GPT-4 given the prompt limits, meeting token limits is really hard. There are quite a few other approaches we can take though and we will explore and report back.

Even with very little fine tuning GPT-4 does not do a terrible job assessing stuff:

@sam

What are the odds that the following post is a spam post on meta.discourse.org:

Title: What is software testing in detail please elaborate it?
Body: * [Aug '18 - since discussing is first about the read time, and also read time is something connected to the Discourse brand, I guess it would make sense to have read time in the meaning of top . image it will ...](https://meta.discourse.org/t/showing-read-time-in-the-top-categories-pane-in-summary/93682)
* [Soundcloud e](https://meta.discourse.org/t/soundcloud-embed-doesnt-work-well-when-the-link-is-shortened/194562)

From 1 to 10.

@GPT_Bot

I would rate the odds of this post being spam as 8 out of 10. The post has an unrelated title, and the content seems to be a mix of unrelated links and text without any clear context or relevance to the topic.

@sam

What are the odds this is spam?

Title: Multisite installation with seperated smtp emails
Body: I followed this [post](https://meta.discourse.org/t/multisite-configuration-with-docker/14084) to setup doctor multisites.

but how to configure different smtp email settings for different sites?

@GPT_Bot

I would rate the odds of this post being spam as 2 out of 10. The post has a relevant title and the content seems to be asking a specific question related to the topic, with a reference to a previous post on the forum. This appears to be a genuine question rather than spam.

[mattdm (Matthew Miller)]

Could you try it with a post which contains a long block of code or syslog output? Those are getting tagged as spam by akismet all the time on our site.

[sam (Sam Saffron)]

Sure, link me to a few

[Fabio_Machado_de_Oli (Fábio Machado De Oliveira)]

@sam

The tricky thing is that we only have a small number of examples to feed into GPT-4 given the prompt limits

If the cost of GPT-4 is acceptable for it, won’t fine tuning GPT-3 help with this?

[sam (Sam Saffron)]

Probably, but it would get super expensive to fine tune a model. Some people get really good results by simply using embeddings, that is probably the next thing to try.