I’ve been playing around with another so-called AI program that creates pictures based on a verbal prompt (wombo.art). While it’s absolutely terrible at mid-level structural stuff, like making sure a person has a head and two legs, it’s pretty good with colors and textures and with overall layout, so many pictures are “hits” on a gestalt, impressionistic level.
Look at these pictures and try to guess (approximately) the verbal prompts that produced them. They’re all pretty straightforward, more “Washington crossing the Delaware” than “chicken dancing with a jellyfish on a giant chessboard made of citrus rinds.”
Picture 1:
Picture 3:
Picture 4:
Guess in the comments. I’ll reveal the answer if anyone gets reasonably close.
16 comments:
Picture 2: The Last Supper
Picture 3: Orgy at the Playboy mansion
Picture 4: Motorcycle gang
Picture 5: American Presidents in a graveyard
It looks to me like various levels of hell.
1. Robin hood and his merry men.
2. Council at five cell
3. Techno party at a castle
4. Motorcycles in the burbs.
5. Trump returns from the dead.
2. Council with Gandalf and elves.
Very impressive!
1. Robin Hood and his Merry Men (bull’s-eye for HomeStadter)
3. Party at the Playboy Mansion (a_probst)
4. Hells Angels motorcycle gang (a_probst)
No one’s got 2 yet.
The guesses for 5 are close but not close enough. Hint; The program put Trump in the picture because it misunderstood the prompt.
2. Knights of the Round Table?
2. King Arthur and the Knights of the Round Table (ben)
That leaves 5 . . .
Be assured I typed my guesses out into notepad and pasted them here before reading the seven extant responses, lest I be influenced.
My impressions are:
1 Robin Hood and His Merry Men
2 King Arthur and the Knights of the Rond Table
3 People Getting Stoned at a Nocturnal Pool Party
4 Hell's Angels Visit Sesame Street
5 Donald Trump Attends Joe Biden's Funeral and Delivers a Soliloquy to a Floating Disembodied Skull
Number 5 remains unsolved. The inclusion of Donald Trump was a mistake on the program's part, and a mistake the nature of which readers of my Tarot blog may be more likely to guess.
5. The Emperor scrying his orb.
Really? I think I still like my answer to # 5 better than yours, William -- and I don't even know what yours is yet!
PS programs don't make mistakes, only users do!!! ;-)
#5 Trumpet of judgment and the dead
@Poppop, programs may not make mistakes but programmers certainly do!
#5 The trump of Judgement XX
And we have a winner!
5. The dead rising from the grave at the last trump
Well, sure, that works too, I suppose.
(@Ben -- as a bit of a coding hound myself, I must raise an unhandled exception to that remark. From the point of view of the CPU, programmers are infallible. When they say "jump", the CPU asks "which memory address"? Rather, programmers are adept at finding new requirements the users never knew they even had, for example getting creative results to a problem. Why, that's the entire basis of AI I think.)
So now you need to have it render the entire Tarot deck perhaps? Although of course final trump is probably most topical for 2022 anyway.
@Poppop nice. I was thinking of the training data used by those who built the machine learning model used by the website. It appears to have biased the model away from brass instruments and Tarot cards in favor of POTUS 45. As here, that leads to behavior unexpected by the end user who has a musical or mystical use case in mind.
Post a Comment