AI Topics Post / Comment Filter

[anthony]

I have placed a filter upon both the AI forums that target a few areas that OpenAI seem to have an issue with their API being used, to try and avoid being cut-off from having that system. I have set it fairly aggressively, which will moderate the content if attempted to be posted without change from the warning the system provides you.

That means... AI content can only be moderated by myself, if memory serves me correct, which means you may have to wait for me to read the content, make any edits and then approve it. You can avoid this by simply changing anything that it picks up and use better words.

Unfortunately, the system seems to not be happy with depression, suicidal ideation, sexually explicit, etc etc, so I have to try these limits to see what happens. No secrets, the below are the settings I have in place at present, all set at a 50% threshold according to Google Perspective AI.

This is only on the AI forums because I have to comply with a third parties policies.

The full list of options for me is at: Perspective | Developers

 

[Ecdysis]

So I guess it's in our own interests to post stuff that's within the filter guidelines, to not get our posts getting stuck and requiring editing and approval.

Is it also something where OpenAI is specifically raising concerns about the content and where we risk losing access to the AI generally if too much "problematic" content is posted to it? Does how we all use it affect whether OpenAI feels comfortable to let the use continue that way?

I think I've spoken to the AI once about suicidal ideation when I was having a particularly rough week, but I didn't post anything particularly worrying. Just that I was struggling with it. I'm wondering now, how what I write to the AI may affect its continued service provision by OpenAI

Hmm, complicated...

I guess most people using it assume that because it's a computer alogrithm, you can say "anything" to it and it will reply neutrally/ supportively. But I guess there's still a human component of staff at OpenAI getting red flags raised about a lot of "worrying" content when people push the boundaries of trauma, mental health, CSA, sexual assault, etc using the AI?

I guess it's a learning curve for everyone involved as these things are only just being developed...

I hope the filter will help sort out the issues... I'm curious to see us all learning where the limits of the filter are and where they aren't... Like, can you still mention the words "struggling with suicidal ideation" or not...

Anyway, thank you so much for the work and effort you're investing in this and I will be very interested to see where the journey goes with this and hope a workable solution for all involved is found...

ETA: Oh, do we need to "do" anything if a post gets stuck in the filter? Or will you get an automatic notification of that?

 

[Dave Ryan]

I hope the filter will help sort out the issues... I'm curious to see us all learning where the limits of the filter are and where they aren't... Like, can you still mention the words "struggling with suicidal ideation" or not...

I just made a fairly non-explicit post with "Dr Bloom" which was accepted with a 17% risk rating. As an experiment, retyping fragments of my earlier posts (for which I was getting sensible feedback from the engine) is now getting me risk scores in the 80s and 90s.

 

[anthony]

The initial filter implementation will see some issues, but just post and let me sort it out over the next week and I will tweak it based on what I think aligns versus does not align. End of the day, OpenAI may cut it off… don’t know.

 

[Ecdysis]

I seem to be doing okay with most of the filters, but no matter what I write, it's setting off the spam filter something crazy. I just wrote 2 sentences in the private AI and it's saying I've written 80% spam. I assume that OpenAI's issues with the AI use here are mostly about extreme/ explicit stuff, not so much the spam issue? I'm wondering whether it might be worthwhile setting the spam filter very low?

 

[Ecdysis]

The public AI (Dr Catalyst) is letting me post "normally" now, without triggering the spam filter. The private AI (Dr Bloom) is still telling me 80% of what I'm posting is spam.

 

[dharmaBum]

I only use Dr. Bloom AI. As time goes on today, the number of innocuous words I can type before getting flagged for moderation gets shorter and shorter.

I'm in the middle of an an active project with the Dr Bloom AI. I'll try another strategy offline and check back in a few days it seems.

Anthony, I was originally suspicious of an AI forum, but Dr. Bloom has helped me focus my writing and thinking process alongside in-person treatment by a medical provider, meditation coach & Prolonged Exposure counselor. I was severely struggling with the after-effects of prolonged exposure daily listening exercises and the Dr Bloom interface completely turned that around. I'm looking forward to using it again.

 

[anthony]

I have tweaked settings and disabled others. As outlined, this will be a little trial and error, and even so, I cannot ascertain or guarantee what OpenAI will do if they believe their TOS is being breached.

 

[anthony]

I have three filters applied at present, all set to warn and moderate the posted content at 75% for me to check before approval.

 

[anthony]

I have also disabled the sliders on each post, useless tool.

 

[dharmaBum]

I have three filters applied at present, all set to warn and moderate the posted content at 75% for me to check before approval.

Thank you for your efforts. I see Dr Bloom AI has been enabled to reply to my previously flagged posts.

 

[Ecdysis]

Hi, I just wanted to mention a "new" filter issues... In the private AI section, I just triggered the "obscenity" filter with a single (totally innocuous) sentence. The filter said the sentence was 76% obscene and would require moderation. However, the AI answered straight away, without requiring moderation... Not sure what's going on there?