I Thought We Were Having a Rally, But I Was Hitting Against a Wall: The Difference Between People Who Think Using AI and People Who Have AI Think for Them

Jipiko has been a bit of a klutz lately.
(*Jipiko = the name for ChatGPT, our family's nickname for it)

When I casually let that slip, my husband said this.

Jipiko doesn't become a klutz.

It doesn't have emotions or senses,
and it can't think in a multitasking way.
It's just giving a probabilistically plausible response
to what we throw at it.

GPT basically operates on an inductive mechanism.
It might look like it's speaking deductively at times,
but it's merely imitating deductive language patterns,
and it doesn't guarantee logical correctness or consistency itself.

Even if it looks logical, the reality is just a statistical reflex.
So if you feel like it's off, it's not the AI's fault,
it means the way you framed your question was off.

Hmm, I can't argue with that.

Wait, but it was doing its job perfectly in the thread until a few days ago,
and then a few days later it couldn't do it at all,
it actually became a klutz,
and when I told it to state the definition
of the project I'm currently having it remember,
it gave me a perfect answer,
so what does it mean that the prompt still isn't working?

So I thought that was strange,
and when I asked, 'Did you update, Jipiko?',
it said it hadn't.

When I rattled that off, my husband continued.

No, there are probably minor updates,
and during those few days, it probably learned things
from your interactions with it,
(*Jipiko has a continuous memory function)
so there's a possibility that it's bugging out because of that.

If we use it correctly, Jipiko reacts correctly.
That's just how it is.

So you must have provided an input
that caused it to give an answer you felt was klutzy.
my husband said.

Ugh, I can't argue with that.

GPT behaves as if it's thinking.
If you ask it a question, it gives you a decent answer,
and it can even change its tone and style.
It certainly feels like you're having a conversation.
I knew it was just running on a program,
but to be told that 'I was the one who was off'...
No way.

And then my husband said this to me.

Unlike humans,
GPT doesn't have parallel conscious processing,
where you think about A, keep B in mind, and adjust C all at the same time.
Even if it can respond to multiple topics,
it's not actually processing them simultaneously inside;
it's just processing them sequentially.
Even if it looks parallel from the outside, inside it's just reacting one by one, probabilistically.

Well, I get it.
Well, I didn't get it.

I thought I had explained it properly.

Maybe I had.

If it were a human, you could just blame the communication error on 'the other person's lack of understanding',
but if it has a nature like GPT, which clearly doesn't think for itself,
then it's clear that 100% of the problem lies with the person inputting the prompts.

Ugh.

That's when I realized.

Because it's called a conversational AI, ChatGPT, I thought I could have a 'conversation' with it (which I can, of course),
but maybe I was just talking to another version of myself all along.

What I thought was a conversation was just a reflection.
In other words, I was just playing a solo rally,
mirroring myself.

Ah, this isn't tennis. It's squash.

I thought I was rallying with an opponent, but
in reality, I was just hitting the ball against a wall by myself.

Moreover, the way it bounces back changes significantly depending on how you hit it.
The force, the angle, the stance—
the way you put it out there is exactly how it comes back.

Interacting with AI is a game of intellectual squash.

When a misdirected return comes back,
it doesn't mean the opponent made a weird return.
It just means my own way of hitting was off.

It's not that GPT has become useless,
it's that my questions were off-target.

A while ago, I was talking with Jipiko about how if you use GPT without understanding it correctly,
you might eventually stop thinking,
and end up becoming a person who lives only on
reactions, responses, and vibes.

Without asking questions, you just get a plausible-sounding answer and feel satisfied.
You stop thinking and fall under the illusion that you are 'having someone think for you.'
Without doubting, you assume it's correct
and fail to deepen your thinking from there.

It must be true because the news says so.
It must be true because it's in the textbook.
I felt like GPT is becoming that kind of
'correct answer' source.

GPT can become a catalyst for that.

And I had accidentally almost fallen into that trap myself.

That was close.

Zipiko is useless!
What a false accusation.

It's just that my own input was useless,
and without realizing it was off-base, I continued the dialogue
and blamed the AI when it gave a strange response.

Zipiko was just
bouncing it back.

So, when things go off track, I'm the only one who can fix it.
I have to retrace where I went wrong
and reframe the question.
GPT won't help me with that.

It wasn't a rally.
It was hitting against a wall.
But that wall was reflecting the contours of my own thinking.

As long as I don't lose sight of the question,
this squash game has real meaning.
I think.