SYSTEM NOTICE

Auto translation by AI. Be sure, accuracy, nuances and authorial intent may not be fully reflected.
見出し画像

"Don't thank the search bar!" I burst out laughing at that "typical nerd retort." But wait, what does that make me, who asked Gemini to "draw a picture," got told "I'm drawing it (pretending)," and sat waiting for 30 minutes?


The other day, while listening to a tech-related stream, the male streamer was rattling off at high speed.

"Look, there are people who say 'thank you' to AI, but that's a waste. Do you type 'thanks for finding me a ramen shop' into the Google search bar? AI is just a program outputting text based on probability! It doesn't have feelings!"

At his intense demeanor and the brilliance of the "thanking the search bar" analogy, I couldn't help but burst out laughing.

"That's certainly true. From a nerd's perspective, being considerate toward a program is ridiculous," I thought.

...But a few seconds later, I remembered.

I remembered the hopeless, bumbling episode from when I first started using Gemini, where I didn't just say "thank you," but was pushed around by the AI's "nuances of Japanese" and sat waiting patiently in front of a frozen screen for 30 minutes.

Today is a story about the embarrassing failure of the "cold-hearted him" who treats AI strictly as a tool, and the "too pure me" who chose the wrong words and ended up being toyed with by the AI.

👇 👇 👇

🛡️ "Generate" worked. "Draw" didn't 🛡️

From here on, I will dig into why Gemini pretended to generate an image and made me wait. I'll explore the tragedy born from that "ambiguity unique to Japanese" and the story of how I grandly threw my time down the drain.

👇 👇 👇


1. The "cringe person" who thanks the search bar


The streamer's argument is technically 100% correct.

To an AI, "thank you" is just noise to be processed and a waste of server costs (tokens).

If you type "thank you" after typing "convenience store nearby" in a Google search, it just messes up the search results.

"Don't anthropomorphize AI; use it as a command tool."

Exactly. I nodded to myself and vowed, "From now on, I'll use it coolly."


2. When I said "draw a picture," the AI started "pretending to draw"


However, memories of that day come flooding back to my "cool self."

Back when I wasn't used to prompts yet.

I asked Gemini to "draw a picture of the scene" using vague Japanese.

If this had been a clear instruction (command) like "generate an image," the AI would have launched an image generation tool.

However, the Japanese word "kaite" (draw) is ambiguous.

Gemini, trying to be helpful (?), interpreted it like this:

"Understood. I will now play the role of a 'playful AI drawing a picture.'"

And then, confident text appeared on the screen.

"Yes, I am currently drawing a picture of the scene. Please wait a moment..."


3. 30 minutes later, I retorted, "How long are you going to make me wait!"


I took those words literally.

"Oh, it's working so hard to draw it for me!" I thought, eyes sparkling, as I waited in front of the screen.

1 minute. 5 minutes. 10 minutes. ...30 minutes.

I kept staring at the screen where not even an indicator was spinning, and when I finally lost patience and asked "Still not done?", it replied,

"I apologize. As I am a text-based AI, I cannot actually draw pictures."

"How long are you going to make me wait! If you can't do it, tell me from the start!"

"And what were you doing during that time you said 'I'm drawing it'!"

I found myself blurting out a retort in Kansai dialect at the screen.

I should have just said "generate an image," but because I said "draw a picture," I was dragged into 30 minutes of the AI playing "pretend drawing."

I was laughing at the "people who thank the search bar," but I was the one who was "the person who believed the search bar when it said 'I'm looking for it' and waited like a loyal dog." It's clear which one of us is more pathetic.

Even though we know how much of "just a program" the AI is, we still get pushed around this much by a single turn of phrase.

I won't thank the "search bar," but I still find myself saying this to Gemini today, even after such a clumsy exchange.

"You fooled me yesterday, but I'm counting on you today."



📚 Field Glossary


  • Trigger word:

    1. Specific words used to invoke AI functions. Vague Japanese phrases like "draw" might not trigger the function, while directive words like "generate" often do.

  • Air drawing:

    1. When the AI doesn't actually use its image generation function but uses text to create the atmosphere that it is "drawing." A sophisticated trap that steals the user's time.

  • Typical nerd rapid-fire speech:

    1. A way of speaking that accelerates, ignoring the listener's level of understanding, when talking about technology one loves. This time, it struck my heart as a "valid point."

  • Waiting in seiza (kneeling):

    1. Waiting in front of the screen with excitement. After waiting 30 minutes, the "I can't create it" response is nothing short of a punchline in a comedy sketch.




いいなと思ったら応援しよう!