Comment by MgB2 3 months ago
The prompt didn't exactly describe Indiana Jones though. It left a lot of freedom for the model to make the "archeologist" e.g. female, Asian, put them in a different time period, have them wear a different kind of hat etc.
It didn't though, it just spat out what is basically a 1:1 copy of some Indiana Jones promo shoot. No where did the prompt ask for it to look like Harrison Ford.
If you take the perspective of all the possible responses to the request, then it is overfit because it only returns a non-generalized response.
But if you look at it from the perspective that there is only one example to learn, from it is maybe not over it.
> If you show the model values for y, but they are 2 in 99% of all cases, it’s likely going to yield 2 when asked about the value of y
That's not overfitting. That's either just correct or underfitting (if we say it's never returning anything but 2)!
Overfitting is where the model matches the training data too closely and has inferred a complex relationship using too many variables where there is really just noise.
The nice thing about humans is that not every single human being read almost every content present on the Internet. So yeah, a certain group of people would draw or think of Indiana Jones with that prompt, but not everyone. Maybe we will have different models with different trainings/settings that permits this kind of freedom, although I doubt it will be the commercial ones.
I didn't think it. I imagined a cartoonish chubby character in typical tan safari gear with a like-colored round explorer hat and swinging a whip like a lion tamer. He is mustachioed, light skin, and bespectacled. And I am well familiar with Dr. Jones.
Is HN the whole world? Isn't an AI model supposed to be global, since it has ingested the whole Internet?
How can you express, in term of AI training, ignoring the existence of something that's widely present in your training data set? if you ask the same question to a 18yo girl in rural Thailand, would she draw Harrison Ford as Indiana Jones? Maybe not. Or maybe she would.
But IMO an AI model must be able to provide a more generic (unbiased?) answer when the prompt wasn't specific enough.
Why should the AI be made to emulate a person naive to extant human society, tropes and customs? That would only make it harder for most people to use.
Maybe it would have some point if you are targetting users in a substantially different social context. In the case, you would design the model to be familiar with their tropes instead. So when they describe a character iconic in their culture, by a few distinguishing characteristics, it would produce that character for them. That's no different at all.
But the concentrations of training data because of human culture/popularity of characters/objects means that if I go and give a random person the same description of a character that the AI got and ask "who am I talking about, what do they look like?" there's a very high likelihood that they'll answer "Indiana Jones".
But... the prompt neither forbade Indiana Jones nor did it describe something that excluded Indiana Jones.
If we were playing Charades, just about anyone would have guessed you were describing Indiana Jones.
If you gave a street artist the same prompt, you'd probably get something similar unless you specified something like "... but something different than Indiana Jones".