Metal Geir Skogul posted:

I don't get it.

Tunicate posted:

https://www.youtube.com/watch?v=r6zZPn-6dPY

A demonstration that a single neural network trained by GAN can generate a variety of images. All images in the same panel coordinate are generated from the same latent variable, and therefore its meaning is preserved over different classes.

End of Shoelace posted:

i would love that aesthetic for a horror game

Bobby Digital posted:

Your Computer posted:

With the low resolution and how fast the images morph I didn't notice it at first, but I paused and uh.. the images are actually pretty disturbing _{(well at least the animal ones)}

End of Shoelace posted:

heres a fun game: play this video at 2x speed and pause anywhere; see what kind of faces you get

Your Computer posted:

oh you know, just typical celeb photos

Hempuli posted:

https://twitter.com/chaingenerator/status/953201759090077696

7c Nickel posted:

https://twitter.com/deepdrumpf/status/728317897412579328

https://twitter.com/DeepDrumpf/status/788918639810478080

quote:

Mis—spun Tales

One of the best ways to see why all the above components are necessary to a story generator is to see how we learned that they were necessary. It is not always obvious how a computer program will actually function while it is still in the planning stages. Important parts of a program are often left out because there was no way to know that they would be needed.

TALE-SPIN, in its early stages, frequently told rather strange stories. These “mistakes” caused many re-definitions in the original program. Since this process of “mistakes” followed by new theory is characteristic of AI programs in general it is worthwhile to look at these “mistakes” and consider what had to be done to fix them. (The output of the original stories has been simplified for ease of reading.)

***** 1 ******
One day Joe Bear was hungry. He asked his friend Irving Bird where some honey was. Irving told him there was a beehive in the oak tree. Joe threatened to hit Irving if he didn’t tell him where some honey was.

Joe has not understood that Irving really has answered his question, albeit indirectly. Lesson: answers to questions can take more than one form. You’ve got to know about beehives in order to understand that the answer is acceptable.

*** 2 ***
One day Joe Bear was hungry. He asked his friend Irving Bird where some honey was. Irving told him there was a beehive in the oak tree. Joe walked to the oak tree. He ate the beehive.

Increasing the range of acceptable answers is not enough. You have to know what the answers really mean.

*** 3 ***
In the early days of TALE-SPIN, all the action focused on a single character. Other characters could respond only in very limited ways, as in answering direct questions, for example. There was no concept of one character “noticing” what another character had done. Hence the following story, which was an attempt to produce “The Ant and the Dove”, one of the Aesop fables:

Henry Ant was thirsty. He walked over to the river bank where his good friend Bill Bird was sitting. Henry slipped and fell in the river. He was unable to call for help. He drowned.

That wasn’t supposed to happen. Falling into the river was deliberately introduced to cause the central “problem” of the story. Had Henry been able to call to Bill for help, Bill would have saved him, but I had just added the rule that being in water prevents speech, which seemed reasonable. Since Bill was not asked a direct question, he didn’t notice his friend drowning in the river. “Noticing” is now an inference from change of location, so Bill sees Henry in the river, deduces that Henry’s in danger, and rescues him.

*** 4 ***
Here are some rules that were in TALE-SPIN when the next horror occurred. If A moves B to location C, we can infer not only that B is in location C, but that A is also. If you’re in a river, you want to get out, because you’ll drown if you don’t. If you have legs you might be able to swim out. With wings, you might be able to fly away. With friends, you can ask for help. These sound reasonable. However, when I presented “X fell” as “gravity moved X,” I got this story:

Henry Ant was thirsty. He walked over to the river bank where his good friend Bill Bird was sitting. Henry slipped and fell in the river. Gravity drowned.

Poor gravity had neither legs, wings, nor friends. Now “X fell” is represented with PROPEL, not PTRANS, that is, as “the force gravity applied to X,” and the inference from PROPEL are not the same as for PTRANS.

*** 5 ***
The inclusion of awareness meant that I couldn’t set up the stories that way I used to.

Once upon a time there was a dishonest fox and a vain crow. One day the crow was sitting in his tree, holding a piece of cheese in his mouth. He noticed that he was holding the piece of cheese. He became hungry, and swallowed the cheese. The fox walked over to the crow. The end.

That was supposed to have been “The Fox and the Crow”, of course. The fox was going to trick the crow out of the cheese, but when he got there, there was no cheese. I fixed this by adding the assertion that the crow had eaten recently, so that even when he noticed the cheese, he didn’t become hungry.

*** 6 ***
Before there was much concern in the program about goals, I got this story:

Joe Bear was hungry. He asked Irving Bird where Some honey was. Irving refused to tell him, so Joe offered to bring him a worm if he’d tell him where some honey was. Irving agreed. But Joe didn’t know where any worms were, so he asked Irving, who refused to say. So Joe offered to bring him a worm if he’d tell him where a worm was. Irving agreed. But Joe didn’t know where any worms were, so he asked Irving, who refused to say. So Joe offered to bring him a worm if he’d tell him where a worm was. . .

Lesson: don’t give a character a goal if he or she already has it. Try something else. If there isn’t anything else, then that goal can’t be achieved.

Ariong posted:

I don't understand how that is possible using 1970's computing technology. I don't believe it. Do you have any more information on this project? I can't find any.

Ariong posted:

I don't understand how that is possible using 1970's computing technology. I don't believe it. Do you have any more information on this project? I can't find any.

Guy Mann posted:

It's easy to forget that ELIZA was created in the mid-60s. One of the best arguments against the transhumanist singularity dorks is the simple fact that things like AI and speech recognition have been stagnant for decades even with the exponential growth of computing power and memory.

Noblesse Obliged posted:

Hello
Smithers
You
Are
Really
Good
At
Turning
Me
On

Dumb Lowtax posted:

We put a man on the moon with 1960's computing technology. This is just some string manipulation

Getting logic-based AI to work is largely the same now as it always was. Classic AI was not data-driven so you didn't need terabytes of training data. The big AI winter happened as early as 1984, *after* the big hype about neural networks for solving problems died down.

That winter slowly ended as people realized that there are still all sorts of applications for even the most limited AI (such as the dumber "big data" statistics based ones that are popular today, and that are hungry for as much high-speed input from the internet as they can process). That sort of AI was succeeding in an increasing amount of niche areas, towards ubiquity, and now there's the whole internet full of new opportunities to use it and show it off. There was nowhere for pretty generated images and Gaston lyrics to go where they'd have been quite as appreciated in the 1970's, versus now with twitter. The big difference that you see today is not that there's some giant body of new research everyone knows all about, or some massive code library that took decades for researchers to build up that you now can't build a product without, or even modern computer speeds -- mostly it's the suddenly increased domain of problems being tried.

Also if you want more information about TALESPIN in particular, you can just Google any of the excerpts I quoted above to get full articles. Mostly old ones, so .pdfs without highlightable text.