It can indeed draw a full glass of wine.
The correct prompt just has to be used.
"Can you provide me with an image of wine in a glass filled over the brim depicting how surface tension works?"
!
Isn't a wine glass typically considered to be served "full" at approximately half of its actual physical capacity? Its my understanding that this is done to allow headspace for aromas to be taken in while drinking for proper tasting.
A typical 'pour' of wine is measured in ounces. At least it was when I was bartending. The amount was definitely determined by the size of the glass for the exact reason you mentioned. The exercise in question is whether you could come up with a prompt to get the AI to create an image of a glass filled to the rim. had it correct.
Yes, this is basically what I had in mind. You need a prompt specific enough to overcome the common understanding of what is 'full' for a glass of wine, which what the AI is 'trained' on.
I wonder if it would be easier if you substituted water in place of wine or if its the glass that trips up the AI.
I'm assuming of course that if I asked a bartender to serve me a glass of water (in a wine glass), the bartender would serve the water filled to the rim..
You can get it to create images of filled glasses of nearly anything except for wine. Maybe whiskey too, but I haven't tried it.
For the same reason AI can't draw an analog clock showing an arbitrary time like 5:19. The training data for wine glasses doesn't include examples of full wine glasses because wine is rarely photographed filled beyond a standard serving of wine. Same goes for clocks where only aesthetic clock hand positions are photographed rather than arbitrary times. AI is only able to mimic rather than be truly creative.
This is just one more reason AI is nothing more than a parlor trick. It doesn't think, it isn't aware and it will never out-do a human mind for versatility. AI is just magic to lay-people but real thinkers know better.
https://poal.co/s/AI/780329/024b4f70-a91c-4fd8-b7d5-40ee314428d3#cmnts
It learned about liquid surface tension though.
Openai is a jew owned company...but what else, right...
I once asked an AI image generator to draw me an inner gear. No matter what I tried, it wouldn't do it. I even sent it a photograph of an inner gear and still it wouldn't do it. It generated some pretty interesting looking gear like contraptions though, but not what I wanted.
(post is archived)