Am I the only one wondering if this really even AI?

ChatGPT
1

I can’t be the only one who’s noticed how obtuse and ridiculous this “intelligence” is, can I? Let me go through a few things I’ve noticed…

I once asked AI to create a snippet of code to display a variable in an input field. It was a small piece of code. I pasted it in its spot and it didn’t work. I pasted it back to AI and it told me my code wasn’t designed to perform the function I wanted. Crazy.

I have given the AI countless blocks of text to try to get it to help me with my job, just so I don’t have to read the text myself. I’ll ask it to just give me the names of people mentioned and it misses people. I’ll ask it to reorganize the text in various ways and it consistently leaves out entire chunks of important text. Now, I can write a piece of code in 5 minutes that can return a piece of text that you type into a field, but something we call Artificial Intelligence can’t? This one in particular really make me doubt the intelligence of it. How can any one in any field use an AI as it is advertised to help you with your task if it can’t even properly reference and return text to you? That’s like a basic function you kind of expect from any tool. How much of a calculator would a calculator be if it regularly left out numbers from its calculations?

I’ve given it countless pieces of my code to ask it to check for syntax or help me debug, and it usually returns the code to you and will literally say “This is your updated code correcting syntax…” but it leaves out entire chunks of the code, or empties the contents of a function and tells you “your function logic goes here”.

I’ve asked it to generate countless images for me and have noted that they always seem to follow the same sort of patterns which tells me it isn’t really creatively or intelligently drawing you an image. Try asking it for an image and specifying that it not include any words or language in the image. See how often you get an image with words in it.

They call it machine learning, but try giving it a set of instructions and then asking it do things expecting that it will follow those instructions, see how often it actually works. It doesn’t learn at all. It can’t even follow an instruction correctly. A child can do that.

Then I find myself arguing with it, which is insane and I really hope I’m not the only person who does that without thinking about it. Am I? Please say no!

2 Likes
2

GPT-4 is not human.

In essence, it’s just a sequence prediction machine. It doesn’t think, in the conventional sense. It doesn’t understand, in the conventional sense.

However, it has some amazing emergent properties that you can harness if you help it overcome its shortcomings. That sometimes requires you to spend a considerable amount of time trying to understand it.

You can almost think of it as trying to get someone with a specific set of neurodevelopmental disorders to accomplish your tasks. Successfully getting them to work first requires you to understand their deficiencies and how to overcome them.

Arguing with the machine is pointless in almost all cases.

That is accurate. You’re getting a trained model - it’s “done learning” - so to speak, until the next version comes out. Trying to “teach it” in the conventional sense doesn’t really work. Every chat is a completely blank slate.

You’re definitely not the only one. I understand it can be frustrating, especially if you expect it to do something it can’t.

but you need to remember - you’re arguing with a piece of electrified rock. As it stands now, it will not budge, and forget everything you said as soon as you open a new chat. It’s extremely pointless, unless you find it cathartic.

You can’t change the model. The only thing you can change is how you interact with it. Using things like Chain of Thought, iterative processing, restricting usage to narrow context tasks… you can try to become a better prompter.

Or you can wait for better models. It looks like better models will have an easier time understanding what you mean - and have a bigger arsenal of tools they can use - so that it requires less skill to get them to do what you want them to do.

1 Like
3

If this is true:
“it has some amazing emergent properties that you can harness if you help it overcome its shortcomings” then why all the other text?

4

not sure what you mean :thinking:

5

Maybe you’re right. I am new to using AI so it’s likely there’s a lot I don’t know. Any pointers?

1 Like
6

The biggest thing I would say is to try to apply some sort of empathy when interacting with the model. If you keep in mind that you can’t change it, and the only thing you can change is your approach to interacting with it, you should be good. :slight_smile:

1 Like