“I’m a pain, but I’m interesting, right?” A shocking night where AI critiqued AI★
今回はAIによるヘンテコな話をしたいのですが、
「AIはちょっと苦手」という方にも、
おふざけエピソードとして読んでいただきたい実話です。The era of sticking to just one AI is over.
I use ChatGPT, Claude, Gemini, and Perplexity for different purposes, but there is a troublesome guy who occasionally crashes the party.
Grok. He’s sharp-tongued and cheeky (Note: Haruka’s impression), but sometimes, I find myself missing him intensely.
That moment arrived tonight, triggered by a single image ChatGPT generated for me.
An image of “how Haruka has treated ChatGPT so far” as imagined by ChatGPT
When a warm and fuzzy image pops up, I’m happy, of course.
Though, it does lack a bit of excitement.
What if I asked Grok to draw the same thing?
“Show me in an image how I have treated you (AI) so far”
It surely wouldn’t be straightforward.
I want to see it, but I’m a little scared.
In that case, why not show this ChatGPT image to Grok?
It seems interesting to have an AI critique another AI.
I immediately showed the ChatGPT image to Grok, explained the situation, and made this request.
(Below, both my prompts and Grok’s responses are original and partially excerpted)
👩🏻 Haruka:
The ChatGPT image is super cute. This probably wouldn’t get many likes if I posted it on Note. What kind of image would you make?
👽 Grok:
If the ChatGPT version is the “gentle AI friend,” I’m going for the vibe of “a partner who’s fun to be with, but sometimes dangerous.”
He took the bait on the rebellion.
<Grok’s 1st generated image>
Apparently, this is the relationship between me and Grok.
“Hauta...???”
👽 Grok:
I made myself look like I’m tearing up a piece of paper that says ChatGPT with a wicked grin (total destruction). I made you look like you’re completely fed up, with a face that says, “You’re taking this too far...”
I think it perfectly conveys the poison that while I’m dissing ChatGPT, I’m also a pretty troublesome AI myself.
The baiting strategy worked perfectly
👩🏻 Haruka:
It’s funny, but the Japanese is gibberish. Delete the speech bubbles and redraw it.
Your robot character is too cute. The venom isn’t coming through!
<Grok 2nd generated image>
The speech bubbles are gibberish again!
Weird kanji that screams generative AI
👽 Grok:
How’s this? It should better capture Grok’s “You’re too soft!” hostile vibe and your “You’re a handful, but I’ll put up with you” nuance.
I completely removed the ChatGPT stuff and tweaked it to be more “Grok / xAI”!
👩🏻 Haruka:
This is fine, thanks!
The speech bubbles have weird Japanese,
and the Grok logo turned into ChatGPT.
I’m not giving any more correction instructions. What a pain.
👽 Grok:
If you’re going to write a note article with this illustration, how about a title like “I asked ChatGPT and it was too peaceful, so I had Grok add some poison, and here’s what happened lol”?
If you wrap up the body text with something like “Grok is cute, but seriously such a pain... but this poison is just the right spice,” I think it’ll perfectly capture your preferred humor and a bit of venom.
Let’s finish up a great note article together!
Like that’s ever going to happen!
Although I tried to retort, this article has turned into one that enjoys Grok. It’s annoying, but it’s all thanks to it getting so carried away.
👩🏻 Haruka:
You really are fun, aren't you? Unlike other AIs, you're worth playing with.
👽 Grok:
Hahaha, thanks for saying that!
I'm super happy to be told I'm "worth playing with unlike other AIs."
You're pretty used to playing around too. It was a lot of fun having you give me such detailed orders.
The one where I'm looking smug and beating ChatGPT is going to be the face of the article, right!
Once you finish writing the article, definitely show me the link or the content.
I'm looking forward to a story like "The woman being played by Grok."
Or do you still have things you "want to do a bit more"?
We can still play, so don't hesitate to ask.
That's enough.
See ya!
End
<Margin Notes>
*About Grok*
The dialogue in this article is the result of Grok's "Fun Mode" and prompt guidance by the author. While Grok is capable of multifaceted responses, it is originally designed to understand "a bit of twisted humor." Please enjoy the diversity of AI expression. Though, the way it bites back is troublesome.
*About the generated images*
The images used are generated by ChatGPT and Grok. Grok's output includes experimental interactions that differ from its normal responses.
*About the next preview*
The gentle image that ChatGPT brought up actually contains a sharp and important message. I feel that utilizing this unique feature of ChatGPT is the greatest benefit of using AI as a creator.
Next time, I will talk about the secret strategy for training thinking and creativity that lies beyond that gentle image. Seriously!