To me, who screams 'The AI just doesn't get it!': The secret to getting along was 'subtraction'

When you're dealing with AI, don't you ever have those moments where you want to stomp your feet and shout, 'Why don't you understand!?' or 'Read the room, will you!?'

I... have had that happen more times than I can count (laughs). At first, I'd get angry, thinking, 'What a clueless thing!', and then I'd try to be overprotective by writing out long lists of instructions, thinking, 'Alright, I'll just explain everything!'

Previously, when I posted the 'realizations' part of this article, I received a lot of feedback. To everyone who read it, thank you so much! However, I felt strongly that 'there must be more concrete things I can share' and 'I want to pass on the practical tips I learned from those failures,' so I have rewritten this with significantly upgraded content.

From here on out, this is a complete set designed to let you use all my failures as your stepping stones—from secret dialogues with my partner Chappy (AI), to 'magic rephrasing techniques' you can use starting tomorrow, and even a practical workshop to turn your AI into a 'master painter.' (Please read this as my perspective at the time of writing.)

Q1: Is there a difference between talking casually and talking in an organized way?

MeiMei:
I'll ask you honestly,
is there really that much of a difference between
talking casually and talking in an organized way?

MeiMei:
Wait—is that just pretending to know?

I thought, 'I see.'

You're not reading my mind, you're just making a selection.
Once it was put that way, the magical feeling suddenly vanished.
In my first month, I was writing everything down.
Cute, but mature, but strong, but delicate,
in a dark forest, but with light shining through—.
As a result, what came out was 'something average that had a little bit of everything.'
Looking at that, I thought, 'You just won't listen to me.'

But now I understand.
It wasn't that it wasn't listening,
it was that I couldn't indicate the priorities and proportions.

Q2: Is it true that the longer the prompt, the better?

MeiMei:
Are longer prompts better?
I feel like I get my point across better if I write a lot.

It doesn't mean longer is better.
I thought that this is the same as when I talk to people.

The more important something is, the more I end up talking about it.
Because I want to convey it properly,
I explain this and that, and before I know it, the other person is overloaded with information.
I remember the troubled look on their face.
I see. So AI is the same...

But how can I get the AI to understand this worldview inside my head?
When I think about that, the words inevitably increase.
That's why it tends to get long.

MeiMei:
So, you're saying it's better to make it shorter.

I see.
Based on that logic, is the length of the prompt the length of my hesitation?
But sometimes, if I decide things too abruptly and concisely, it doesn't go well for some reason.
That's a problem in itself.

Q3: Why does it 'blow up' when I set things too strictly?

MeiMei:
When you set conditions strictly and then make a big change, it suddenly gets weird, doesn't it? Like the Japanese becomes buggy or something.

MeiMei:
The safe side?

If you tell a person 'this is important and that is important,' they'll be troubled too.
If you tell them 'this is the most important thing,' it might be easier to act.
Maybe it's the same thing... I wonder???
Hmm, this is starting to get deep.

Q4: What kind of distance is just right?

MeiMei:
Is it okay to be rough at first?
Or is it better to solidify it to some extent before throwing it at you?

Approaching in stages.

In my first month, I was trying to get the right answer in one go. It was still free back then.
But now, I can stop at 'I've gotten this close today.'
I'm using the paid version now, but the actual number of outputs I generate until I reach a level where I can give it an 'OK' hasn't changed much from when it was free.

You don't have to complete everything.
You just need to take one step forward.

Once I started thinking that way, the amount of dialogue naturally decreased.
It decreased, but I feel like the accuracy has improved.

Having written this far, I feel like I've finally begun to understand the dialogue above a little.
The AI isn't 'guessing' anything.

From the words I've provided,
it's just choosing the one that makes the most sense.
If it's vague, it leans toward the average.
If the priorities are clear, it leans toward those.
It doesn't mean it's understood because I wrote a lot, nor does it mean it's sloppy because it's short.

In the end, it's about
whether you can properly decide for yourself 'what you want to show the most.'
If you have a 'decisive point' within yourself, you'll reach the correct answer quickly.
Maybe that's all there is to it.

Oh, I see.
That's all it takes.

What if the correct answer is blurry?
There's a way to handle that too.
Since it's off-topic for today, I'll save it for next time.

Extra Bonus Track

This time, I'm intentionally not including any images.

Instead, please try this with the AI you are using
by setting it to 'generate a new image' mode.
Alright, here we go!

First, try giving the instruction:
'Please draw a white cat.'
You were able to generate an image, right?

Next, try giving the same instruction:
'A white shorthair cat. One eye is blue.
Looking a bit sleepy in the morning light.
The background is a kitchen.
Please generate a new original image.'

How is it?

There must be a difference.
There should be.
There is, right???
Assuming there is... Ahem.

That difference isn't because the AI suddenly got smarter, but because
you organized your thoughts.
If you do this, you're practically halfway to generating the image.

*Please try this in a new chat.
If you continue in the same window, the AI might already remember your preferences.

See?
Doesn't it make you think
that's all it takes?

The answer is probably
right there on your device, isn't it?

See?