Hacker News Reader

Comment by pyentropy | Hacker News Reader

I don't get it? Most of the "limitation" examples are on par with humans, both linguistically and artistically.

So much so that it feels like a DALL-E ad :)

...you know a lot of people who can't count past four? Who don't understand the concept of negation? Who don't understand what a modern supercart aisle looks like? Who lose track of elements in a description after four or five? Who can't get that when you ask for a "monk and a skeleton," you don't mean that the monk is also a skeleton? Who literally don't know what letters are?

A lot of people can't subitize past four. Counting is another matter: it's a symbolic, deliberate, step-by-step activity.

Arguably most neural networks don't do symbolic, step-by-step reasoning "natively", and maybe human beings don't either. But they can be told to, and then the same network becomes much smarter: https://arxiv.org/abs/2201.11903

> Who don't understand the concept of negation?

I teach students of the first year of the university, and plenty of them have problems with reading comprehension, in particular to understand sentences with negations.

This, pretty much. I sometimes feel we might be underestimating AI because most people on HN are well above average.

What comes to mind is all those cakes with bad inscriptions. First Google result: https://images.app.goo.gl/JZtBVCwmCFfcmV6F8

Yeah, a lot of people seem to have problems interpreting instructions for drawings.

Reply on news.ycombinator.com