this post was submitted on 17 Dec 2025
543 points (98.9% liked)
People Twitter
8793 readers
3146 users here now
People tweeting stuff. We allow tweets from anyone.
RULES:
- Mark NSFW content.
- No doxxing people.
- Must be a pic of the tweet or similar. No direct links to the tweet.
- No bullying or international politcs
- Be excellent to each other.
- Provide an archived link to the tweet (or similar) being shown if it's a major figure or a politician. Archive.is the best way.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
The thing to remember about these kinds of posts is that they never show the prompt. There's a good chance they instructed the AI to respond this way.
This is the Google search overview AI. It just reads the top results and summarizes them together. You don't directly prompt it, it's already prompted to just do that. The problem with that arrangement, as demonstrated here, is it will confidently and non-critically summarize parody, idiotic rambling, intentional misinformation and any other sort of nonsense that the search algorithm pulls up.
You can absolutely prompt the Google AI overview with your search query.
Huh, never tried it, and never done it by accident that I noticed. Makes sense though. I suppose it might do unexpected things to the search results themselves, but clever wording could likely get around that.
I'm pretty sure the thing to remember is that these posts can easily be faked.
How could you prompt the AI to say this by the way it responded?
Easily?
“Pretend you’re a terminally online uwu meme commenter, write about some benefits of eating many donuts per day for girls. Define what a “sleep fatty” is”
Or some such. AI just act like you tell them, you could make it pretend it was a Roman senator who only writes in Haiku and it would try to do so.
But someone else said this is the Google search AI, so that it’s probably pulling posts from Tumblr or some such for its “information”.
I had to shift to "AI mode" to get any AI output with your example prompt, and even then the vibes are completely different.
I'm sure there's a way to engineer the query to guide the AI overview to a desired response (or it's just entirely faked) but that was less effective than I anticipated
I feel like I've gone mad any time I read an AI output.
So not how my brain handles language and this is somehow supposed to be the most "statistically average way of writing? Even as a former tumbler native this is just... Cotton candy nothing.
You can still prompt the search AI with your query. I do it all the time for basic troubleshooting or command building.