In case you’ve ever wondered how to use a menstrual cup, don’t ask AI. They might give you a nightmare illustration.
But wait! There’s more! I decided to ask the Google AI to explain the diagram. It didn’t see any problem.
AI Overview
• This 3D medical animation illustrates a medical-grade silicone menstrual cup inserted into the vaginal canal to collect rather than absorb menstrual flow.
• The visualization highlights the proper sagittal view placement, emphasizing a comfortable position below the cervix and angled towards the sacrum.
• It serves as a reusable, eco-friendly alternative to traditional period products like pads and tampons, capable of being worn for 8–12 hours.
Now you know. Just punch the cup into the bladder and through your cervix to completely replace your uterus.
David Cronenberg himself couldn’t imagine this level of body horror.
What was your input query?
Proximity to that uvula doesn’t look comfortable. Maybe it’s a burst appendix; that would explain the hole.
What type of alien is this uterus in? Gray? Reptilian?
Icky, like so much of human biology, but not as bad as what you routinely see on The Pitt..
There seems to be entirely too much anatomy in the same position, in addition to the cervix cup borg horror and that plastic camera at the top. Spy ovaries? Digestive tract? Is that a kidney just floating around like a balloon?
It needs to work on those labeling lines. None of those bits are located within the vagina.
Wondering if this reflects a fundamental limitation of AI, an inability to “think” and reconstruct structures in 3-D?
Limitations of the query matter too, rorschach.
Obs no ‘thinking’, because it’s a LLM.
And whoever put the query in to generate the image also got it wrong.
(Still, I do like the verterbrae and the sacrum in the background. ;)
If you had an AI generate 100 images on a well worded query what would result? Pick the best 5 and worst 5. I always wonder if AI image horror shows are cherry picked from the worst results pile, though if bad stuff winds up in textbooks or research publications that’s not good.
I’ve had decent results in Canva with AI, though I have to cherry pick the best results after tweaking my instructions to get by the obtuseness of the language model. And sometimes the slop is more a feature than bug if it adds a subtle tincture of the uncanny or something humorously odd.
But asking AI to generate an image demonstrating insertion of a menstrual cup seems doomed to failure from the start. What is the “bladder” the image is pointing to?
What? No. I would iterate a given image, not select from a sample and be done.
Well, yes. With your approach, it would be.
Clearly this Abominable Intelligence fad is the future of humanity. It couldn’t possibly be a stupid waste of time with no value, oh no…
I use Google search a lot.
Many times a day at the very least. Already today I’ve used it to look up the latest membership declines of the SBC.
Lately in the last week or two, Google Search has gone way downhill.
I always check what the AI part at the top comes up with. Sometimes it is wildly wrong.
I put a query in the search box asking what percentage of Boomers voted for Kamala Harris.
It told me that it couldn’t answer because the 2024 election hadn’t happened yet.
I found Google Search easier and more reliable before Google started trying to insert their AI into everything.
That diagram is Horrifying! But, it’s not important. AI doesn’t care how many humans (women, men, children) it kills with its crap info and phony, destructive job displacement.
And, don’t you dare mention how many thousands of peoples’ lives are being destroyed by the berserk proliferation of data centers. The nashville zoo says it has animals that will probably die from the noise, power shortages, etc. of a data center to be built next door.
And whoever put the query in to generate the image also got it wrong.
Okay…care to tell us how one could get such a query “right?”
I’m not surprised Gemini found no problem with it. Current “state of the art” LLMs, with the notable exception of Qwen, are notoriously horrible at understanding images. They don’t actually get the images at all, just 1 token for each 16×16 block of a low resolution version of the image, often without any position information.