Feeding the zoo animals (Site Stuff)

[BridgetYourFortyDaysAreUp]

Sorry about the weird thread title; it's deliberate.

I thought everyone on FWR should be aware of this thread on Site Stuff (not everyone looks at Site Stuff).

It concerns what you put in your draft posts before you hit "Post."

More specifically, for a good summary, see

LemonDog45 at 22.11 yesterday

https://www.mumsnet.com/talk/site_stuff/5585556-odd-ai-generated-thread-title-options-in-site-stuff?page=9

And JustineMumsnet response at 16.29 today

https://www.mumsnet.com/talk/site_stuff/5585556-odd-ai-generated-thread-title-options-in-site-stuff?page=10

Make of that what you will. Justine says Mumsnet opted out of allowing OpenAI to use our draft posts. Would you believe OpenAI if they said, "ok, thanks for the info, we won't use the draft posts"?

This thread may get deleted, but, for now, I would avoid typing/C&Ping anything identifying or personal into your draft posts, even if you intend to delete it before posting, because it looks as though OpenAI may be scraping your draft posts as it generates titles and even possible responses. Before you actually post.

Might explain some of the weird, relentless posts we've had on FWR lately. Or maybe there really are just quite a few bored people around at the moment.

I'm still amazed they are so hand wavy about it, when everyone else has to label their AI it seems Mumsnet have given themselves their own rules to live by.

Thanks. I always mean to check site stuff and dont. Apparently there was a suggestion the FWR should only post about trans people or something (without letting anyone on FWR know they had gone behind out backs).

But interestingly, as I did wonder, I noticed that new threads no longer have the option to have a title generated for you.

Whilst it is clear that the forum than MNHQ provides is hugely important for women, there seems to be some sort of real life understanding, empathy or whatever with how women use the site.

Unless this was another example where a geek thought technology will always be better, rather then MNHQ thinking women are competent.

Sorry for terrible typos in previous post.

My head is still full of cold two weeks after I stopped having a cold.

You and everyone else. 🌹Commiserations.

I didn't notice any typos. I'm too busy worrying I'll be shunned.

Why would you be shunned? You are only sharing info from another part of the site.

Might explain some of the weird, relentless posts we've had on FWR lately. Or maybe there really are just quite a few bored people around at the moment.

I do think there is something else going on, or not just one thing, but that's for another thread.

Thanks for sharing this info.

Yes I’ve been following this on the site stuff thread, and posted my own weird AI experience on the thread. I think your advice is pertinent.

Yes, saw you there, Eresh. It's an odd thing, and the risk of personal info being scraped before posting is just not something I had even remotely considered. I know nothing about AI except that it exists and everyone is using it.

It seems like a rather cavalier UE failure to me, the IT people may not have thought about how people use the posting box to draft what they want to say and get their thoughts straight, and aren’t going to be comfortable with it being visible as they are drafting it and before they’ve decided to send. Particularly if they edit it to remove personal info or sensitive stuff.

Also I think people don’t think of OpenAI as a particularly trustworthy company, unlike how they may see MN.

Do you think things would improved if there was a little icon next to the AI generated thread titles?

I do quite like the idea that all AI content gets marked with a little icon so we can know. Not sure it matters that much in thread titles, though.

The AI generated title option is no longer there.

If it's broken and they're trying to fix it, that's not surprising.

Yes they said they were suspending until they’ve decided how to make it opt in and transparent.

Well that's a positive in terms of site stuff working.

But if I remember rightly at the time it suddenly appeared many pointed out the flaws with it.

I think the real concern here is that OpenAI may be scraping data from posters' drafts. Before they post, even if they delete that data before they post. Once the AI kicks in to generate a title (this function which has temporarily been disabled), AI has access to everything. It doesn't need to wait for you to post your final post.

Especially risky for those who C&P from other sources before editing, then posting.

Even if it is opt in for the OP it should be marked AI for the reader.

I get that. I don't have anything intelligent to add to that debate at the moment though and was just interested in this side bar. Sorry if it's derailed things

Thanks OP for highlighting the AI access to drafts. That's a massive breach of trust as far as I'm concerned. I've been considering leaving MN and deleting my account for some months, mainly because of the number of male posters, both TRAs and those purporting to be sex realist. This is the shove I needed.

Not saying this will re-assure anyone as for all I know by now AI tools have formed a union, but according to Google AI there is a difference between OpenAI individuals can use via ChatGPT and their commercial model for businesses like mumsnet.

1. The Public App (What people think of)
If you log into the ChatGPT Website as an individual and type a prompt, OpenAI does save it. By default, it stores your conversations in an account history and reserves the right to use that text to train future AI models. This is the "vast database" people fear. 1, 2, 3]

2. The Business Setup (What Mumsnet uses)
When Mumsnet connects to OpenAI behind the scenes, they are using OpenAI's commercial API. OpenAI's legal terms for this type of traffic are entirely different: 1]

• No Training Data: OpenAI explicitly promises that any data sent through a business API is never used to train ChatGPT or other AI models. 1]
• Temporary Cache, Not a Database: Instead of a permanent database, the text is sent to OpenAI's server, processed in milliseconds to generate the title options, and then held in a temporary safety log for up to 30 days purely to check for automated abuse (like malware or illegal text) before being deleted. 1, 2]

(Sorry this has probably already been explained on site stuff.)

As a postscript Google AI added

• Lack of Consent: The tool sends your draft text to OpenAI automatically while you type before you have even hit the "Post" button. If you write a deeply personal draft, change your mind, and delete it, that text has already left Mumsnet's servers and visited OpenAI. 1]
• The Irony of the Lawsuit: Mumsnet is currently in a high-profile legal battle with OpenAI, suing them for historically "scraping" Mumsnet's public forums to build ChatGPT. Users find it incredibly hypocritical that Mumsnet HQ is suing OpenAI for data harvesting out the front door, while feeding users' unpublished private drafts directly to OpenAI out the back door. 1, 2, 3]

And

It revealed that AI prompts were pre-programmed for replies

The glitched text contained backend instructions and templates for generating automated replies to the "Am I Being Unreasonable" (AIBU) forum. Users immediately caught on that Mumsnet wasn't just testing a simple title maker; the code suggested they were building or experimenting with a framework to have AI generate actual posts and responses on the site. 1]

MNHQ's handling of the situation made users even more suspicious. When posters first flagged the weird "Response Schema" text, MNHQ initially denied it, claiming it hadn't come from them. Shortly after, they had to backtrack and admit it was their internal tech. While they strictly maintained that they were not using AI to write actual forum replies, the leaked title text proved the backend capability was actively being worked on. 1]

And finished with

If there is one thing MNHQ should have learned over the last two decades, it is that you cannot sneak anything past the Mumsnet user base.

I find it quite disturbing to thinks that MNHQ might have thought it wanted to generate fake AIBU threads, let alone fake posts responses.

Although if the last is true might explain why there are now so many posts on FWR which only seem to be about telling FWR it doesn't have the right to exist (for being unkind)!

Just bumping this in case some haven't seen it.

To be clear: this is a concern, but MNHQ has said that OpenAI has not been given permission to do this. You can make up your own minds what to think about it, and if you trust OpenAI.

Please see the Site Stuff links. It sets it out pretty clearly.

LemonDog45 at 22.11 yesterday
https://www.mumsnet.com/talk/sitestuff/5585556-odd-ai-generated-thread-title-options-in-site-stuff?page=9
And JustineMumsnet response at 16.29 today
https://www.mumsnet.com/talk/sitestuff/5585556-odd-ai-generated-thread-title-options-in-site-stuff?page=10

I would avoid typing/C&Ping anything identifying or personal into your draft posts, even if you intend to delete it before posting, because it looks as though OpenAI may be scraping your draft posts as it generates titles and even possible responses. Before you actually post.

ed. random formatting