Posts being automatically flagged

I guess we have a new feature that I need to study. Posts are being auto-flagged and hidden. An innocuous post I just made in the Win11 tablets thread was automatically flagged. Seems like the system is a bit touchy. The guidelines seem nice (if long!) but it’s not immediately clear what caused the flag. Anybody know the ins and outs of the flagging feature?

Edit: it gets better, I made a small edit to the list of tablets (adding the HP Fold to the OP), and then THAT post got flagged. Scrolling down I see

Flagged by the AI NSFW filter. I mean those curvy tablets are definitely sexy, but not over the line… right? OK, off to the forum settings I go.

Wow, the AI section has a LOT of options. The folks at OpenAI (right?) are working hard to make AI useful in our daily lives. Needs some tweaking though.

Current settings were:

I’m going to dial those up a bit.


So given how annoying it is to get an innocent post flagged, and given that I think we have deleted a total of… two? posts in the past year, maybe we should dial these settings wayyyy up, or disable the feature? Thoughts?

Can we maybe turn it off, and do test runs on a single post once a true NSFW one finally shows up. E.g. try to repost that a few times in an admin-only thread “AI test thread” to find a reasonable detection level? As is, it adds too much friction (ooohh… sexy).


And one more comment: the message “the community feels (the post) is offensive, …” seems too harsh given that it might be due to a prude AI based detector. So maybe “the community or its automated sex bot post flagging system finds the post potentially offensive, …” etc.

The crossed out part was a joke, but the rest could stay.

1 Like

Yeah, I noticed one of Dale’s posts was hidden because of this. When I saw the post later, I was pretty confused about the message =P

1 Like

We have community guidelines? Who knew?

Our decency is why people flock to this forum.

Aww… I was hoping flock would be censored :slight_smile:

1 Like

The AI feels like it is part of our community. Clearly it wasn’t trained well enough on the subject of so many of the posts here =P


Lets see if “get the flock out of here” gets flagged… :laughing:

edit: answer; Nope

I noticed this last night. The current post is the edited and “approved” unhidden post. All I did was delete the starting image for my post, which was the Picard face palm gif I’ve used here periodically.

1 Like

Yeah it looks like that part of the filter was dialed down a bit. Feature update or a fix by @Hifihedgehog ? Either way, it’s an improvement!

So I wonder if the AI image scanning is acting like the one with our internal chat system ?

With it, any image that has what looks like hair in it that doesn’t have a human face associated with it in the same overall image gets flagged.

It took us months to figure that one out :upside_down_face:

Hey, and the Picard gif has a face but no hair in it. I think we’re on to something here. :joy:


I think the Picard image is a “trigger” issue because it signifies disapproval, and surely we can’t stigmatize any subject with disapproval…

ha ha !

Seriously though, on further thought I’'ll bet it triggered a possible content copyright violation as the filters seem to be pretty zealous on that score.


I had raised the thresholds two days ago and was monitoring the results to see how it would react, and it introduced some hallucinations, so I had reverted them myself before you had even messaged me privately. No action required on your end, but thanks for being proactive!


So I have found my new go to substitute word for a bad one.

The system seems to have no problem with things like “shut the flock up” either :slight_smile:


Oh cool. So is this something offered by Discourse for free, or is it a subscription service?

Pretty interesting how quickly potentially useful AI tools are being released into the wild. And longer term, also interesting that people could use these tools to "moderate the internet’, e.g. ask their helper AI to remove topics or types of comments from anything they read. Customized bubbles for everyone!

Or “burst bubbles?”

It can cost as can everything with Discourse but we self host so I self host their Docker-based LLMs for toxicity, NSFW and Sentiment within our environment so it is free. The only part we have any charged services is the Discourse AI summarization and bot. That operates solely through OpenAI for ChatGPT and we are charged for that from OpenAI, and Discourse has no involvement there.


My days are numbered.