I’ve moved this to UX since I think this is a more appropriate category for this conversation.
We’re actively working on composer redesigns, so I’ll share this feedback with the design team working on that project.
We’re considering how to better surface other AI features like AI titles, categorization, and tagging because we share your concerns that these useful features are not being surfaced as well as they could be.
I would like suggested layout here. Mainly because I’m so bad with English and it fixes, not just grammatical error, but it proposes better? wording and sentences too (except it is totallu sure right form is a user).
Meaning that is good position if
an user has rights to use AI helper
a forum is global and using one working language
But with forums that are more regional ones it is unnecessary. My forum is for finns who can Finnish. They don’t need that (and there is dialect question too).
Native speakers may understand value of proofreading
I would like a suggested layout here. Mainly because I’m not great so bad with English and it fixes fixes, not just grammatical errorserror, but also it proposes better better? wording and sentences too (except when it’s totally unsure of the it is totallu sure right form, such as form is a user).
Meaning that it’s a is good position if
a an user has the rights to use an AI helper
a forum is global and uses using one working language
But with forums that are more regional, it’s regional ones it is unnecessary. My forum is for Finns finns who can speak Finnish. They don’t need that (and there’s also the there is dialect question to considertoo).
Once you learn CTRLALTP, there is no turning back. However, learning it is not easy, as it depends on a user taking a random walk. Many will not find it.
I do wonder if there is some sort of configuration here from a user/site operator’s perspective:
Teach users about proofreading (for the first N times a user posts, teach them about proofreading - prompt to proofread prior to posting)
Always proofread my stuff (user setting for integrating into “Reply”)
Also… long term, I can see us doing background proofreading as you type, so we can prompt people while composing that something is off.
Yes, I think this is probably the best option but in the meantime:
We could do something similar to automatic image captions. Where we initially prompt the user once with a popup that asks, “Would you like to automatically proofread posts” and if they select “yes”, all their future posts will be proofread when hitting reply.
It works with typos, and perhaps with wrong prepositions, but never with the wrong order of words in a sentence.
That is more or less a matter of the level of proofreading. And I’m a bit worried that background work would water down the whole idea of AI? Plus… OpenAI or another pay-per-token user wouldn’t be totally happy, because that would require constant traffic between a forum and OpenAI, right?
But sure, if we are talking now about how Meta would use it with self-hosted AI, then things are different. Or am I again missing some important background knowledge or even the point of everything? Wouldn’t be the first time
This 100%. For our Problem Solving Community, the title is moderated in about half of all topics because it’s the critical piece of information when browsing search results in list view.
For anyone looking for something similar, I use the following prompt and then provide the body of the topic to the LLM:
For all subsequent messages, we are writing a title for a discourse topic where:
- title is less than 20 words
- title is very concise and matter of fact
- title is a question
- title is from the point of view of the author
- title should be concise and accurate
- typically should be in the form of "how do i..?" or "will...?" or "should...?"
ready?
This converts something like “Export code review findings” into “How do I export a list of open findings from Git code reviews for detailed tracking?” which provides far more context about the topic
To further emphasise this, I’ve used Discourse everyday for 3 years now, and still didn’t know that little gem
This is a good idea. It addresses the most important dependency “User needs to know proofreading exists in Discourse”
I suspect proof as you type would be tied to a self hosted LLM, or for enterprise customers.
How did you use the proofread text feature in the past weeks without seeing
which is, in my opinion, a great way to learn that there is a keyboard shortcut.
I understand the need to surface this feature but I’m reluctant to add it as a second primary button.
I like this idea quite a bit. A just-in-time “did you know about proofreading” message when someone is composing a post or reply (perhaps over a certain character length?) seems like a great way to educate members on this feature.
@chapoi sketched out this idea which I’m hoping she doesn’t mind me sharing
