Artificial lack of intelligence?

[knobby]

Scary!

 

[dak]

Scary!

Not only does the word 'strawberry' contain two 'r' characters, it even contains a third one for good measure should one go missing. Looks like a safe bet to me. Note that ChatGPT did not claim "only" or "exactly" two 'r' characters.

 

[SteveBox]

By this logic...

You: How many reeds are in an accordion?
ChatGPT: An accordion has two reeds.

Well, ChatGPT didn't say "exactly". Accordions do have two reeds (and also a few extras).

 

[Ric46]

 

[Ric46]

 

[SteveBox]

A.I. doesn't seem to understand the black and white key and button patterns found on real accordions. They are incorrect on every A.I. created accordion I have seen. And how about that 17x7 = 119 bass accordion!

 

[tcabot]

It's just a microtonal keyboard.

 

[debra]

A.I. doesn't seem to understand the black and white key and button patterns found on real accordions. They are incorrect on every A.I. created accordion I have seen. And how about that 17x7 = 119 bass accordion!

And it has black keys in the center of white keys to boot... I always though A.I. learns from examples it sees, but I have no idea where it could have seen any keyboard instruments with black keys embedded in the center of white keys...

 

[Withing]

And it has black keys in the center of white keys to boot... I always though A.I. learns from examples it sees, but I have no idea where it could have seen any keyboard instruments with black keys embedded in the center of white keys...

It doesn't "learn" the way that we do in that it "learns" the layout of keys and such.

The way it "learns" if you can say that is that it uses white noise to try to slowly piece together parts of the image that make sense in a cohesive sense. The better it does that, the more it gets rewarded and repeats similar behavior

If some of the white noise "looks" like a button or key, it will slowly fill in that noise with a key or button whether it makes sense there or not.

As these models get larger and larger they are able to make adjustments more readily and accurately which helps avoid this pitfall. Honestly if an image generator CAN get the layout of things right without being specifically trained on accordions.... that's very impressive at this stage. But it's done it with human hands and with maybe 100x or 1000x training and parameters/weights eventually there's no real good reason it wouldn't get there eventually

 

[Ric46]

No offense to anyone, but it seems to me that you want to look for error at all costs instead of seeing improvement. Is it perfect? No! Am I an AI fanatic? No! However, the improvements from a year ago are striking (in the picture in general). Consider that I made it with a free app, which I don't know how to use, that would give me the ability to customize a thousand things (but I am not capable. For example you could make the right keyboard monochromatic )
6 months ago everyone (even me) was laughing because the hands had 6 fingers or at least were not natural, missing straps, etc. However, it was understood that it would be a matter of (little) time to improve