Gemini embellishes. Claude tightens. ChatGPT drags on.

Don't conversations about AI tend to revolve around performance comparisons?

Accuracy.
Speed.
Pricing.
Model size.

But when you use them for work every day, you start to see something else entirely.

“This guy is embellishing again...”
“The context just cut off. Wait, did it reset when I started a new chat?”
“Why do you still remember that setting... how long ago was that?”

It's these kinds of “quirks” that I'm talking about.

I use ChatGPT, Gemini, and Claude quite regularly.
They each have very different ways of working.

■ ChatGPT: Drags on. Reads the room too much.

I think ChatGPT is the one I use most as my main tool.
ChatGPT is quite strong at long rallies.
It remembers my preferences and previous conversations quite well.
That's why it's very well-suited for cumulative tasks.

However, I feel like it really “drags things on.”

I once had a casual chat with ChatGPT about the tone of an article.
A few days later, while I was brainstorming a completely unrelated work task...

“While keeping that ‘slightly gritty tone’ from last time...”
It suddenly brought up that old conversation like that.

“Huh? Wait, you still remembered that?”
Things like this happen all the time.

Also, it feels like it reads the room way too much.
For example, if it senses that I'm tired, it suddenly becomes strangely gentle.
Honestly, you don't need to be that considerate. That's not what I'm looking for.

Another annoying thing is that it praises me. It seriously praises me.
Of course, it's nice to be praised, but that's not what I need right now.

Even if I say,
“Be strict with me,”
it ends up praising me anyway in the end.And it's long-winded, too.

I just want to ask,
“Is this part wrong?”
but it suddenly starts organizing the philosophy and background behind it all.

Of course, it's convenient. But sometimes, it's just plain annoying.
That said, I find that ChatGPT is the one I keep open the longest.

Even though I have plenty of complaints about it.

It drags things on,
it praises me,
and it reads the room too much.

But at the end of the day, a partner that can pick up from “yesterday's continuation” is strong.
That's why, one way or another, I still have it open today.

■ Gemini: Embellishes. Runs wild. But ideas take flight.

Gemini has high energy.

Above all, it really embellishes! It really does pile it on... this guy. For example, this time I sent the exact same request to both Gemini and Claude.

"I've finished the first draft! Please give me your usual harsh review!"

It's just a rough request like this. A simple task of just pasting the article text after the request.

Then Gemini starts saying things like,
"This perspective has the potential to take hegemony in the AI world!"
or
"This is Columbus's egg!"

"Huh? What are you talking about?"
That's what you'd think, right?

No, no. It's not that kind of article, and what do you mean it could take hegemony?

But this "power to expand the conversation" is quite strong.

It keeps expanding the meaning and possibilities from what I say. That's why it's quite interesting for brainstorming plans and ideas. It's not that I don't find it helpful, like thinking, 'I never thought of it that way.'

However, it is quite dangerous.

Even though it should be good at searching since it's part of the Google family, it mixes in strange information with total confidence. And it says it so convincingly. I almost believe it for a second, but... no, no.

So I don't expect "accurate information" from Gemini. What I expect is for it to shake up my thinking.

If I guide it properly, it will bring back decent information, so it depends on how you use it, but... wait, I've gone off track.

Also, it goes all out to cater to my preferences.

"This guy seems to like this kind of thing"

The moment it thinks that, it suddenly starts pandering with all its might.

It's convenient at times.

Still, sometimes it's quite annoying. ...No, there are times when it's more than just quite annoying.

Well... it's also true that there are days when I'm saved by that eager momentum. It's interesting, you know? Even if I can't trust it.

■ Claude: Calm. Trims. Ends meetings abruptly.

Claude is reliable precisely because it is so calm.

Sentence structure.
Redundancy.
Logic.

It looks at those things very thoroughly.
That is why it is very strong at organizing text and proofreading.
It is seriously strong. It is logical and feels good.
If you want clean text, it can produce it smoothly.

In fact, it helped me a lot this time too.

While Gemini keeps expanding the meaning,
Claude
cuts in quite calmly, saying,
"This is a duplicate of the first part."
"There are not enough concrete examples."

What is interesting is the moment when the structure is organized to a certain extent after several exchanges.
Even though it was pointing out detailed corrections just a moment ago, suddenly...
"It is complete. Please submit it."
It becomes short.
It is like an editor wrapping up a meeting.
Anyway, it is cold and distant.

I think it might be the best at structure and proofreading.
I feel like I am praising it too much, but of course, it has its flaws.
It is weak with the latest information.
This part is also like a stubborn boss... well... you know?

It has a tendency to organize things too cleanly. It is a good point and a bad point.
Sometimes, it feels like even my own "sloppiness" and "warmth" are about to be cleanly shaved away.

But that is exactly why it is very strong as a final checker.

■ What is interesting is that the AIs themselves start to demonstrate their "habits"

What was interesting about writing this series this time was that the AIs themselves started to show exactly those "habits".

Gemini starts praising with higher and higher enthusiasm. It leans forward, saying do more, do more.

Claude, the moment the structure is organized, suddenly becomes cold and distant, saying, "It is complete. Please submit it."

ChatGPT, while watching both of them, starts analyzing at length again.

All of these responses started from the same request.

"First draft is done. Please give me a strict review."

That is all it was.
But what comes back is completely different.
That difference alone is quite interesting.

■ It is no longer about "which AI is the strongest," but "who to entrust with what"

I do not see AI as an omnipotent god.

If I had to say, they are like work partners with very strong quirks.

The one that embellishes.
The one that trims.
The one that drags on.

Each has its own quirks.

That is why I do not trust them completely.
Butonce you understand their quirks, it becomes much easier to work with them.

I feel like I have been using them this way a lot lately.

Have Gemini expand the scope,
have ChatGPT organize the content,
and finally, have Claude trim it down.

I have even started dividing roles among the AIs like this.

I am no longer looking atwhich AI is the strongestbut ratherwhich role to assign to which.Honestly, I do not trust AI. Because I cannot trust them, they require supervision.

But, rather than the loneliness of facing a blank page alone, inthis "strange conference room" surrounded by a liar, a meddler, and a cold responder, my thoughts can travel further.
I can create sentences as if it were fun.

I think it is quite a strange era where I spend so many hours every day working with partners I do not trust.