ChatGPT's Deep Research went rogue. The story of how my note market research ended up at the National Diet Library

I only wanted to research the note article market, but
ChatGPT's Deep Research flew off into space 🚀

The National Diet Library appeared in the references.

No, that's not right.

I didn't want to conduct a national-level investigation.

What I wanted to do was market research for creating note articles.

What I should write next.
How to divide free and paid articles.
What kind of themes might reach readers.

It was research to look into those areas.

But Deep Research took it seriously.

It took it too seriously.

Before I knew it, I was at the National Diet Library.

e-Stat.
e-Gov.
EDINET.
J-STAGE.

????

Documents that looked like blueprints for Japanese deep research appeared.

I just wanted to write a note.

I didn't want to draft a national budget.

At first, I thought ChatGPT had gone rogue.

Well, maybe it is half-rogue (lol).

But when I looked closely at the content that came out, it wasn't wrong as research.

In fact, it was quite accurate.

To put the contents of the PDF briefly, it was like this.

First, don't go looking for the answer right away.
First, refine the question into a verifiable format.
What exactly are you investigating?
What is the scope of the subject?
Finally, write it by separating facts, inferences, and unknowns.

No, it's correct.

It's too correct.

But I just wanted to know what I should write about next on note.

Yet what came back was a world where an investigator puts on white gloves and enters the archives.

It's correct.
It's correct, but it's heavy.

It's like trying to make curry and ending up researching rice distribution using Ministry of Agriculture, Forestry and Fisheries statistics.

Why did this happen?

Honestly, I have no idea myself.

I can only laugh.

I do have a hypothesis, though.

At first, I was passing the prompt as a file.

There is a possibility that Deep Research read it as reference material,
rather than instructions to execute.

So, before executing the research, it started analyzing the prompt itself.

The result: the National Diet Library.

I can't say for sure. But it makes sense if you think about it that way.

Even if it makes sense, I can still only laugh.

So now, when I pass a prompt as a file, I make sure to add one sentence at the beginning.

"This is not reference material, but instructions I want you to execute."

Just doing this has made a big difference.

Before polishing the prompt, that one phrase was more effective.

It feels a bit unsatisfying. But, I have no choice but to laugh.

I think many people have had the experience where asking an AI led to the scope getting way too big.

At that moment, I started doubting how I was handing over the task rather than the AI's performance.

Are you providing that prompt as reference material?

Or are you providing it as an execution instruction?

Whether to go to the National Diet Library or return to note market research.

That fork in the road is, surprisingly, right there in the chat box.

You can read it for free via X repost👇
I've slowly restarted on X, so I'd be super happy if you followed me🤗

Site map & a brief self-introduction of mine👇️

#DeepResearch
#ChatGPT
#AIUtilization
#HowToWriteNote
#WritingTechniques
#40sOfficeWorker
#TipsForPosting
#AIandCreation
#NoteManagement
#FailureStory