I asked for, “A caterpillar D4, model 2T, on a farm, pulling a John Deere model 35 combine.”
These are SO bad:
The View From North Central Idaho
Ramblings on explosives, guns, politics, and sex by a redneck Idaho farm boy who became a software engineer living near Seattle.
Well, at least they, sort of, pretty much, got a wheel at each corner.
Everything else looks wrong – including the tread direction on the front wheels on the tall “tractor” in the lower right (not to mention the deformed “Cyclops” lighting arrangement and whatever the software thought rear wheels were supposed to be in the top left image).
Some of us won’t worry much about AI because from examples like this it’s apparent AI cannot dress and feed itself; others of us worry a hell of a lot because AI is starting to look like Automated Democrat Governing.
All I can say is: Sarah Hoyt keeps reminding us to “keep your clothes and weapons where you can find them in the dark” which is useful advice under any circumstances, but “know where the Master Power Switch is” also seems like it may become useful information at some point.
Especially if Democrats suddenly become enamored of AI.
Every D4 I’ve ever seen had tracks, not wheels!
Yes. It’s just one of many problems with this nightmarish art.
i don’t think john deere or cat are in any trouble soon. kinda stupid looking.
Where did Binghamton (Leadbottom, to you of a certain age) get these tractor models, anyway? They have that bizarre Socialist-realism look that so much art from the Soviet Union had in the 1930’s when it was unpatriotic to insist on pay, or food,or shelter, clothing, or even water in order to make you quota of production this month. After all,that’s all less important than creating a new world, a new economic system with equity for all, and a new Soviet Man.
Keep in mind that these AI algorithms are stochastic. The process to construct an algorithm is to take a collection of images and then apply sequentially random images that will maximize the entropy at each step. That process is continued until all that is left is a random image. IOW words you have to destroy the training data. Then you find the inverse of all the sequentially applied random images to construct the generator. Note that this is product of random matrixes.
Then to generate new images you just need to compute the inverse of the product of the random images and multiple by another random image (the seed). There is absolutely nothing intelligent in this algorithm except selection of the training images.
Given the algorithm the best you can expect is the generation of imaginary possible scenes. Mathematically, the goal is to compute some form of a likelihood of a scene given the seed and the training data.
What is important is that this algorithm tells us that our world is a random one of multiple layers of self-similar entities driven by randomness. Friedrich Wilhelm Nietzsche was more right than wrong.
And if you ask where does the randomness come from, look at the three body problem. https://en.wikipedia.org/wiki/Three-body_problem.
There is nothing AI about these algorithms except the name.
The D4 is a series of small tracked vehicles. You might use one to pull agricultural equipment on extremely soft soil or where more traction is needed, but it wouldn’t look like that. I’ve just spent 15 minutes searching the Caterpillar web site, and the only 4-wheeled vehicle I found was a front loader. That was with 4 equal-sized tires, not the big rear wheels, small front wheels that Bing drew.
Conclusion: Bing ignored “caterpillar D4, model 2T” and looked for pictures of farm tractors, of other brands than Caterpillar. Also, I’m not sure if that thing behind the tractor is a combine.
Or perhaps the query was just so wrong Bing could not find any match, and it spitballed the picture. Most combines seem to be self-propelled rather than pulled by tractors. “John Deere 35” only pulls up pictures of conventional wheeled farm tractors. (But John Deere uses letters instead of numbers for their farm tractors.) If I didn’t miss something, my only criticism of Bing Create is that it went ahead and tried to invent something matching less than 50% of the query, rather than stopping and telling you that it can’t find any such things.