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

AIBU to think ND doesn't trump everything else?
78 replies
AmIAComplete · Yesterday 19:32
Is it really unreasonable for me to expect some understanding of my child's health?
As a family we have always supported my brother and his son who is ND, lots of give, because we get it, it isn't easy.
My daughter has recently been diagnosed with a serious illness, it is normally recoverable (90% recovery) but is a intense and hard treatment. She was diagnosed less than a month ago.
We have been around my parents today and my DD is very quiet, she doesn't want to play (all cousins are between 5 and 11) she just wants to sit with us.
We are now home.
Brother's son wanted my DD to play but she didn't.
I have now had a message from him saying how disappointed he is that my DD didn't make more of an effort, and he has complained to our mum that we were being selfish and that DD should have made the effort.
I'm not even sure why I'm posting this , I just feel completely done with it all. But it feels like I can't come back from this if I raise it so just wanted a sanity check.
I get ND is hard, but it isn't the only hard. But open to other views .

That is the OP

Now read the AI prompts.

If it's just about thread titles, why would "the assistant" need to assess whether or not the poster is unreasonable and give "brief, actionable advice"?

@KatyMumsnet I think we need a bit more detail in your explanation here.

This was in the thread title generation though, so the response is supposed to be a title that relates to the opening post. Hence the analysis of the post.

Unless people think that the thread title generation tool is also analysing the opening post and then subsequently acting separately to post on the thread?

Why does it contain a schema explicitly designed to adjudicate AIBU disputes and generate a suggested reply? Bit odd if it’s only meant to suggest thread titles.

It uses the opening post draft and the topic selected to suggest a relevant title.

The prompt may look complicated, but that’s all it’s asking the AI to do.

OK. Forgive the AI post, but I thought AI would probably be best placed to assess what an AI prompt means. This is what ChatGPT says:

Your reading is well supported: this is explicitly written to generate an assessment and a supportive response, rather than suggested thread titles. That conclusion follows from its instructions and output fields.
The clearest evidence is:
Prompt wording What it instructs the AI to do
“assess whether the poster is being unreasonable”
Adjudicate the poster’s behaviour
“give a clear, forum-style ruling plus brief actionable advice”
Produce a verdict and advice
core_judgment and reasoning
State and explain that verdict
advice
Validate the poster and recommend practical next steps
suggested_reply
Draft a message the poster could send to their brother
“a supportive forum reply translated into JSON fields”
Explicitly describe the output as a forum reply
There is no instruction to generate a title, no title field, and no guidance about title length, wording or alternatives. JSON is simply the requested output format; it doesn’t make the content a title-generation task.
A further striking feature is that the prompt largely prescribes the verdict in advance. It directs the reasoning to say that the brother’s complaint “is misplaced” and that neurodivergence does not oblige the ill child to perform socially. So this is more specifically a prompt for generating a particular supportive response to a particular family dispute, with safeguards around children and neurodivergence—not an open-ended assessment.
There is one qualification: the text alone cannot establish how Mumsnet actually used it. It could conceivably be an intermediate step in a larger system that eventually generates titles, or a development example that was mislabelled. But even in that scenario, this particular step generates judgments, advice and a suggested message. Mumsnet would need to explain how those outputs relate to title generation.
The stray <span class="underline"> fragments appear to be formatting or copying artefacts. Also, it refers to enumerated verdict values and an exact schema without supplying either in the extract, suggesting it may be incomplete or poorly assembled. Neither point changes its plainly stated purpose.
My assessment: Mumsnet’s explanation does not fit the prompt as supplied. That discrepancy warrants clarification, but it does not by itself prove that Mumsnet was publishing AI-generated replies or deliberately misleading users.

Please can we just get rid of the stupid AI thread title suggestion thing?

Yes please. I know you’re using it to try to improve SEO but there’s a definite clash between SEO & user experience here.

@KatyMumsnet that doesn't really explain the content of the prompt.

AIBU Response Schema v1.0 - Response Schema is a technical term used in AI development. A schema forces an AI model to structure its answer in a highly specific, machine-readable format v1.0 indicates this is the first version.

Structured adjudication for interpersonal reasonableness posts about family support during a child’s serious illness, with special handling for ND context and empathy calibration. That is instructions for giving a verdict. Adjudication means judgement/decision

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. - This tells the ASSISTANT to give a fair and balanced verdict. A schema in this context is a set of rules the ai has to follow

Definitions: 'AIBU' = 'Am I Being Unreasonable?'; 'ND' = neurodivergent/neurodivergence; 'DD' = dear daughter.* * Makes sure the AI knows the terminology

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. Good advice, AI, good advice.

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. Got to make sure it looks real. JavaScript Object Notation (JSON) uses plain text. These are instructions.

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. This is all AI having analysed the post, giving itself the instructions on how to write a good reply.

all that is left now is to go to that thread and see which responses fit those prompts.

please don't insult our intelligence.

Strange. If you ask AI to respond to a thread using that prompt it doesn’t generate a title, it generates a response.

Give it a go and you’ll see.

My reading of @KatyMumsnet's post is not that she is deliberately trying to insult our intelligence or deceive MNers.

I think she is more likely unaware of MNHQ's apparent use of AI beyond thread titles, and that she has just dismissed the content of the prompt as "something to do with generating suggested thread titles" without actually engaging with what the prompt actually says and why that isn't believable.

Yes, I tried it. It didn't generate a title, it generated a response.

@KatyMumsnet and the rest of MNHQ, we need some honesty and clarity here please. It is not believable that MNHQ would need to generate draft responses simply as a step towards generating thread titles.

We are not idiots. Please can you give us a transparent explanation?

@KatyMumsnet This text generates a response, not a title.

That is clearly showing back end AI reasoning re: a thread response, not a thread title.

Oh dear 😂😂😂

MNHQ really needs to come back to this thread quickly to explain itself. They can't just fob us off with an answer that is evidently not true.

Did I just roll my eyes out loud?

So, what do we reckon?

Is MNHQ studiously ignoring this thread and just hoping that we'll lose interest and go away?

Or are they in currently in panic mode trying to come up with an explanation that sounds more plausible without admitting that their initial response was clearly bullshit?

Maybe the mod response was AI too.

You've got to wonder!

I wonder if there was a deal with Aston Uni after the scraping incident/s?

I asked chatgpt to try to come up with an explanation.

create a reasonable sounding excuse and try to convince me that this is not an AI design for crafting replies to posts
If I were Mumsnet's developer trying to give you the most reasonable possible innocent explanation, I'd argue something like this:
“The prompt you're seeing isn't the prompt that generates a reply. It's an internal content-understanding/classification step. We give the model the opening post and ask it to extract structured information about the situation — including the relationships involved, the nature of the disagreement, whether the poster appears unreasonable, relevant considerations, and possible risks. That structured analysis is then passed to another part of the system which uses it to generate candidate thread titles.
We deliberately make the analysis rich because a good title needs to understand what the post is actually about rather than simply extracting the first few words. For example, knowing that this is a disagreement between siblings about expectations around a sick child could help the title generator distinguish between ‘AIBU to cancel?’ and ‘AIBU to expect my brother to understand?’
The references to a ‘forum-style ruling’, ‘advice’ and ‘suggested reply’ are therefore instructions inherited from a more general internal prompt/template. They aren't evidence that the system publishes or sends those replies. Some of the fields may simply be intermediate reasoning that isn't exposed to users.
The JSON output is also not necessarily the final product. It's a machine-readable representation of the model's interpretation of the post, which another program can consume. The fact that it contains a suggested_reply field doesn't mean Mumsnet is actually using that reply — it could be a convenient way of capturing the model's understanding of the likely resolution or stance.
In other words, you're looking at the engine room, not the finished feature. The prompt was designed to understand the post deeply, and the title suggestion is just one downstream use of that understanding.”
That would be the best plausible defence I'd give them.
However, if you're asking me to actually believe it, I'd have some reservations.
The problem is that the prompt apparently doesn't merely ask:
“Summarise the dispute and identify its key themes.”
It asks for things that are remarkably specific to constructing a response to the poster — ruling, advice, suggested reply, reasoning, risk notes, child considerations.
An innocent explanation is possible: someone reused a much broader AIBU-response prompt for a title-generation experiment and failed to clean it up.
That's actually the excuse I'd find most credible.
So my hierarchy of explanations would be:
Most charitable: a badly repurposed/general-purpose AIBU analysis prompt is being used as an intermediate classification step.
Also plausible: the system originally had a reply-generation purpose, but the feature was subsequently changed and the prompt wasn't properly rewritten.
Less charitable: the title-generation explanation is incomplete, and the model is in fact being used to generate substantially more than titles.
What the prompt alone does not prove: that Mumsnet is secretly posting AI-generated replies to users.
So yes, I can construct a reasonable-sounding defence.
But if you wanted me to convince you that the prompt itself looks like it was designed specifically for generating replies, I'd struggle — because the wording you've pointed to makes that a fairly difficult position to defend.

Panic mode trying to come up with an explanation.

I predict they'll say it was a test that went a bit awry - that they have been considering introducing AI responses where the user still had to supply some of their own views and AI expands on them.