Разбор Discourse 3.4

[Saif (Saif Murtaza)]

Here at Discourse, we’re always striving to make community management easier. That’s why Discourse 3.4 has introduced a more consistent admin experience, a simplified signup and login flow, mobile app improvements, and AI-powered tools to fight spam and help manage your community.

---

This is a companion discussion topic for the original entry at https://blog.discourse.org/2025/02/unpacking-discourse-3-4

[StefanoCecere]

Love you!
and we are already with Discourse 3.5.0.beta1-dev

[darkpixlz (Pyx)]

Many thanks for finally addressing multiple topic drafts! Better late than never

[JimPas]

Everyone at Discourse, both team members & contributors, have been very busy! Thank you for making Discourse… Discourse!

[jordan-violet]

I would love to know if Discourse has any recommendations on different models to use for different use cases.

There are so many foundational models today and over time they’re showing what they are and are not good at.

I’d love to know what models are best for spam, which are best for sentiment, etc. For example, I have gpt-4o and when enabling spam was told I could not use this model with this feature.

[awesomerobot (Kris)]

I think it’s hard to recommend specific models because there’s so much variation from site to site, even within specific tasks (and also budget constraints)… but as far as this goes:

@jordan-violet

I have gpt-4o and when enabling spam was told I could not use this model with this feature.

is this still the case today? I’m able to use 4o for spam on a test site, but if you’re still having issues send us a message and we’ll take a closer look!

[jordan-violet]

@awesomerobot

is this still the case today?

Just went back again to screenshot the error for you and now, of course, it doesn’t come up. Not sure if good or not

@awesomerobot

I think it’s hard to recommend specific models because there’s so much variation from site to site, even within specific tasks (and also budget constraints).

Right, million dollar question. I’m hoping eventually Discourse can come up with recommendations though through model testing with neptune.ai?

[sam (Sam Saffron)]

we did have this documentation topic:

What LLM to use for Discourse AI?

It’s important to understand the needs of you as the community admin and your members when choosing a Large Language Model (LLM) to power Discourse AI features. Several factors may influence your decisions. Performance for use-case. Are you looking for the best-performing model? Performance can vary depending on task, the assessment is based on the model’s ability to generate correct and relevant responses. Context length. The context window is the amount of text a model can “see” and con…

It is very hard to keep it up to date cause every day there are new models released. As a general rule though GPT4-o is good for pretty much anything in AI.

Evals are something we are thinking about and testing, for sure.

[eviltrout (Robin Ward)]

It’s so cool to see progress. That admin interface has come so far! Also, I appreciate the subtle shout out my game got in the blog post

Big to everyone working on Discourse.

[jordan-violet]

Please let me know if you ever want any participation, always happy to help!

[kyle987 (kyle)]

Thanks for the update