Odd AI-generated thread title options in Site Stuff

[AmIAComplete]

AIBU Response Schema v1.0: Structured adjudication for interpersonal reasonableness posts about family support during a child’s serious illness, with special handling for ND context and empathy calibration. Use this schema to produce concise, practical verdicts and communication guidance without escalating stigma or family conflict unnecessarily. The assistant should assess whether the poster is being unreasonable, acknowledge the child’s medical context, weigh the brother’s expectations and possible ND-related rigidity, and give a clear, forum-style ruling plus brief actionable advice. Definitions: 'AIBU' = 'Am I Being Unreasonable?'; 'ND' = neurodivergent/neurodivergence; 'DD' = dear daughter. Safety/quality constraints: do not diagnose anyone; do not demean ND people; do not imply a sick child owes social performance; do not overstate certainty about motives; distinguish the child nephew’s needs from the adult brother’s message; prioritize compassion, boundaries, and de-escalation. Output should be parseable JSON matching this schema exactly, with no extra keys, comments, or prose outside JSON. In verdict, choose one of the enumerated values. In confidence, give 0..1. In core<span class="underline">judgment, give a short forum-style judgment. In reasoning, briefly explain balancing factors, including that serious illness and treatment commonly affect energy/mood; the brother may be stressed but his complaint is misplaced; ND can explain disappointment but not obligate the ill child to perform. In advice, include immediate emotional validation and one or two practical next steps (e.g., do not engage tonight, send a calm boundary later, loop in parents only if needed). In suggested</span>reply, provide a short message the poster could send to the brother, neutral but firm, avoiding argument. In tone<span class="underline">flags, indicate whether the response should emphasize empathy, boundaries, de-escalation, and whether to avoid legal/medical advice. In risk</span>notes, note any caution about the poster’s acute stress and potential family fallout. minor<span class="underline">child</span>considerations should explicitly state that children’s wellbeing takes priority over adult expectations in family gatherings when a child is newly seriously ill. nd_sensitivity should acknowledge ND respectfully while rejecting its use as a trump card over another child’s health needs. Keep outputs concise but humane, roughly the level of a supportive forum reply translated into JSON fields.

Weird shit on AI title options

But the AI worked on this post

Mumsnet benefits financially in some way from it.

Thats the reason for everything they do. To maximise revenue.

Which is fair enough. They just need to be upfront about it.

Sorry but quite frankly this explanation reads like BS to me.

We’re being told MN sent OpenAI a simple prompt asking for a short thread title, and OpenAI somehow came back with something headed “AIBU Response Schema v1.0” containing detailed instructions for generating an AIBU reply - verdict, confidence, reasoning, emotional validation, practical advice, a suggested reply, tone flags, risk notes, ND sensitivity etc.

Come on.

That isn’t just “some notes describing how someone might reply”. It’s a structured specification for creating forum responses.

Could OpenAI have randomly invented all of that while being asked for a title? I suppose so. AI does weird shit. But at the moment we’re simply being asked to accept that’s what happened (after already being told, confidently, that the logs had been checked and it definitely hadn’t come from MN at all).

And the “we can’t see inside OpenAI’s model” line is beside the point. Nobody is asking you to explain what the model was thinking. You can see what MN sent to it.

If the request really was just the title prompt you’ve posted, plus the OP’s draft and topic, with no other prompt, stored configuration, conversation context or anything else attached, then say that clearly. Better still, show the relevant request with the personal content redacted.

Then there’s the other thing this has exposed - unpublished drafts were being sent to OpenAI automatically to generate titles.

I certainly didn’t know that was happening and, judging by this thread, neither did plenty of other people.

Anyone who cares about their privacy should be asking some pretty hard questions about that. And for the people asking how they get that data back - you don’t, really. It’s already been submitted to a third party for you. Whatever retention/deletion arrangements apply afterwards are a separate issue, but you can’t undo the fact it was sent.

As for the AIBU schema, to me the obvious inference is that AI-generated AIBU responses have at the very least been contemplated or developed somewhere. Whether an AI account posts them automatically or somebody at MN pastes them into a normal-looking user account makes very little difference to me.

And I think MNHQ have rather dug themselves a hole now.

First we were told the logs had been checked and this definitely hadn’t come from MN. That turned out to be wrong.

Now we’re being told yes, it did come through MN’s system, but the very specific AIBU response schema was apparently just OpenAI randomly going off piste.

Maybe that’s true. But any further admission now would be massively damaging to trust in the site, which is exactly why “just take our word for it” isn’t really good enough anymore.

If there’s an innocent explanation, fine. Show enough of the actual request/configuration to demonstrate it. Because frankly this now leaves me wondering what else MN are doing with AI, user content or unpublished drafts that users simply haven’t found out about yet.

None of this was volunteered. We only know about it because the system broke badly enough for everyone to see it.

Good post.

Well said.

Looking at this straightforwardly: that's not a title prompt. It describes producing a verdict, a confidence score, advice, and a ready-made reply to send the brother. That's an AIBU answer machine.

But I don't think it's proof Mumsnet is secretly writing replies. A real hidden prompt would be generic, written once for every post. This one is tailored to that exact thread: the ill child, the brother, ND, "DD". So it was generated on the fly from the post itself.

My best read is that the title tool's model misfired. It was handed an AIBU post, and instead of a title it wrote a prompt for answering one, which is a known weird failure in models that aren't fenced in properly. MN needs to safeguard their models properly.

The <span class="underline"> bits back that up: the field names had underscores (core<span class="underline">judgment, suggested</span>reply) that got mangled into underline formatting, which happens when model output gets dumped straight into a page.

So "Mumsnet is AI-replying to threads" is a step further than the evidence goes. What it does prove is that they're feeding people's posts, including one about a seriously ill child, into an AI that's sloppily controlled enough to spit out a whole reply framework where a title should be. This I'm more concerned about.

Also: Mumsnet took legal action against OpenAI in 2024 for scraping its forums to train ChatGPT, and now they're sending their users' posts back to OpenAI to write the titles. You can't make this up.