The Case Where AI and Dragon Quest Were the Same: How to Use Spells Skillfully and a Little Equipment Enhancement

I would like to talk about AI using the analogy of the familiar RPG "Dragon Quest."

Asking AI for an opinion "vaguely" is like casting a spell where you don't know what will happen,"Pulpunte."You get intoxicated by the merely pleasant "Sweet Breath," and end up wiped out while your allies (your own judgment) are dancing.

However, even with the same "opinion," if you ask
specifically,it's a different story.And making it "investigate" is the act of opening a

"treasure chest."—Since the "type" of content is determined, you can assess the risk before opening it.—However, you must not let your guard down with treasure chests either.
Precisely because most are "hits," only those who neglect to check will get bitten by the


Mimic (=hallucination)that occasionally mixes in, and get wiped out.The true nature of "convenient but scary" is not the AI's performance. It's a story about how it lies in your own

"how to use spells" and "equipment."

1 | This one remark turned into an article

Let me tell you about the time I started using AI in earnest.

At first, the situations where I used it for work were quite limited.
Drafting emails, summarizing meeting minutes—that level of "simple tasks."

Only when I stepped in and threw avague"What do you think about this?" did I feel swayed by the hits and misses of the returned answers, feeling that it was "convenient, but somehow scary." It's close to the feeling of consulting a fortune teller.

That feeling started to change once I became conscious of "how to throw" prompts at the AI.
I stopped throwing things vaguely and started asking specifically or making it investigate—the moment I shifted to that, the hits and misses decreased smoothly. And, at one point, I let this slip out to myself.

"Asking AI 'vaguely' is like casting Pulpunte. You don't know what will happen. It's useless."

"But when you ask 'specifically' or make it investigate, it's like the feeling of opening a treasure chest. It's clear whether it's a Mimic or a Metal Slime armor."

After saying it, I retorted to myself.
—Wait, doesn't this hit the essence quite well? lol

This one remark became the backbone of today's article just as it was.
Thinking log technique (making conversation logs with AI an asset) is, in short, this.
Pick up a remark you were going to throw away and structure it.

Let me state this in advance.
"AI is not a wall-hitting partner"is a story about the AI's nature, whether "AI is a reflector or a thinking subject."
"Will AI lose trust?"is a story about "AI degradation." Both have AI as the subject.

That's not the point today.The subject is humans—I will talk about thepattern of proper usewhere the certainty of what comes back changes entirely depending on "how you use" the AI. If this is misaligned, you will misread everything.

2 | To those who don't know Dragon Quest (just 30 seconds)

For those who don't get it even when I say "Pulpunte" or "Mimic," here is the minimum. If you understand this, you can read the rest.

"Dragon Quest" is a national RPG where a hero travels while fighting monsters.
In battle, you fight by casting "spells" (magic). Both attacks and healing are spells.

Among them, there is a unique one calledPulpunte.When you cast it,
what happens is completely random..

There are big hits that wipe out enemies, and big misses where for some reason allies start dancing and waste a turn. It is synonymous with a spell that relies on luck, where you cannot choose the effect.

The other one istreasure chests..
They are placed here and there in dungeons, and when you open them, you get useful items such as weapons, armor, medicinal herbs, and gold.

The contents aremostly "hits.".
However,very occasionally the treasure chest itself is a monster..
This is aMimic..

It's a fairly strong enemy that bites you the moment you open it.
The important thing is here.

When you open a treasure chest, what comes out is either a "useful item" or a "Mimic,"it's only one of those two.The "type" of content is determined.
—I want you to keep this feeling in the corner of your mind and proceed.

For example, in the game, it unfolds like this (both are worst-case patterns).

This "worst-case pattern" happens exactly the same way with AI.
I will look at them in order.

3 | General AI view: Asking AI "vaguely" = Pulpunte (luck game, party wiped out)

Pulpunte is a spell where you don't know what will happen.
Piro-piro-piro♪... it's fine that you cast it, but sometimes it wipes out enemies, and sometimes allies start dancing and waste a turn. It's strong when it's strong. Butyou cannot choose.The important thing here is—the true nature of Pulpunte is not "asking for an opinion" itself.


"Asking vaguely."A blank, round-robin throw like "AI, what do you think about this?" This becomes Pulpunte.

• You don't know what will come back until you open it

• Subjectivity × Black Box × Luck

• If it hits, you're a genius; if it misses, it's just a "plausible-sounding lie"

In other words,the result space is unknown. Both good answers and bad answers exist within the same probability space, and we cannot see the shape of that space.
That's why it's "convenient, but scary."

It's not scary because the AI is weak.
It's because our questions are vague, and we haven't designed what will come out.
(So what happens if you ask specifically? I'll show that later as a "good move.")

4 | Professional AI Perspective: Making it investigate specifically = Treasure chest (Is it a mimic, or a metal slime?)

And the end point of pursuing that "specifically" is not asking for opinions, but "making it investigate." This becomes the act of opening a treasure chest.

You don't know the contents before you open it.
However, the "types of contents" are determined. Medicinal herbs, gold, or liquid metal armor—or at worst, a mimic.

In other words, only "good things" or "traps" will come out. The range is fixed.

"List the main literature on this point and the main counterarguments to it" is the act of opening a treasure chest.

What comes out: literature, existing theory, or consistency?
Mimics (= misinformation/hallucinations) are certainly mixed in.
But if you know they are mixed in, you can verify them.

There is danger.
However, the range is known. That's why you can open it.

Between Palpunte and a treasure chest, if you only look at the level of danger, the treasure chest might be higher.
Mimics will bite you normally, lol.
But people open treasure chests. Because it's not a game of luck.

5 | Essence: Is the result space "unknown" or "known" = "designing" uncertainty

This is the core of today.
The difference between Palpunte and a treasure chest is not a difference in AI performance.

It is the difference in whether the result space is "narrowed down" or not.

If you vaguely dump it, it's unknown (Palpunte).
If you ask specifically or make it investigate, the result space narrows and leans toward known (treasure chest).

Even with the same AI and the same model, it can become either Palpunte or a treasure chest depending on how you frame the question.

"What do you think?" only = probability space is undefined. That's why you can't manage risk.
"Based on the premise of △△, provide pros and cons with evidence" = probability space is defined.
Therefore, it becomes a risk-managed exploration.

You don't "erase" uncertainty.
You "design" uncertainty. This is the watershed.

6 | Therefore, the object of trust is not the AI, but the "exploration protocol"

And, I'll confess.
What I trust is not the AI model in front of me. It's the procedure for opening the treasure chest. I call this an exploration protocol.

• What to ask (ask so that the types become known)

• How to verify (verify assuming mimic contamination)

• Which counterarguments to demand (crush convenient answers)

In short, I've stopped treating AI as a "wizard" and am using it as a "dungeon exploration staff." There is danger (mimics). But the range is known.

Check your equipment yourself. You make the final decision—that set of gear is your exploration protocol.

As long as this protocol is active, the results you get will be generally stable, whether the model is GPT or Claude.
Conversely, if the protocol is sloppy, even the highest-performance model will turn into Chaos.

Honestly, looking for a "reliable AI" is the wrong approach.
What's reliable isn't the AI, but how you open it.

7 | Practical Application: Question Design Determines 90% of Reliability

Conclusion.AI reliability is determined 90% by "question design," not model performance.
Let's try converting Chaos questions into Treasure Chest questions, Dragon Quest style.

═ Battle 1: The Chamber of Chaos (Asking AI for an "Opinion") ═

═ Battle 2: The Chamber of Treasure Chests (Having AI "Investigate") ═

━━━━━━━━━━━━━━━━━━
<Summary>
・For opinions (Chaos), change from vague to specific so you don't get intoxicated and can level up = How to use spells
・For investigations (Treasure Chests), cast Inspect before opening = Equipment/Verification
With these two, AI exploration becomes generally safe.
━━━━━━━━━━━━━━━━━━

Even when dealing with the same AI, your own way of moving can turn a game of chance into a risk-managed exploration.

Battle 1 is "How to use spells (question design)," and Battle 2 is "Equipment/Verification (Inspect)." It's exactly what the second half of the title says.
I'll leave you with a catchphrase.

If you use AI as Chaos, you'll feel anxious. If you use it as a treasure chest, you'll feel excited.
There are dangers. But the scope is known. That's why you can open it.

8 | However, don't let your guard down with treasure chests either (this is the real key)

In the exploration log earlier, the one thing I glossed over was the "Battle 2 Bad Move Route." That is actually the scariest part. I'll put it into words properly.

I'll confess honestly.
I've done this many times in Dragon Quest.

Treasure chests are almost always "hits."
The probability of a Mimic appearing is low.

So, eventually, you get careless and start opening them one after another without even checking.
Then, you get bitten by the occasional mimic, and your guard-down party gets wiped out.

The exact same thing happens with AI.
Most AI search results are 'mostly correct'.

So you end up letting your guard down and using them without verifying.
And then, you get bitten later by the one hallucination—a mimic—that was mixed in.
'Wait, where did you get those numbers?' you're asked in a meeting, and your whole argument is wiped out.

The scary thing about treasure chests isn't that there are many mimics.
It's that you let your guard down because they 'only appear occasionally.'

The scary thing about AI isn't that it makes many mistakes.
It's that you stop checking because it's 'mostly correct.'

So, what do you do? In Dragon Quest, the answer is clear.

• Strengthen your gear(Prepare background knowledge and a checking system before opening)

• Improve your party composition(Add another AI or human eye to the mix = Multi-AI/Review)

• And then—Cast 'Inpas'

Inpas is a spell to identify the contents of a treasure chest 'before' you open it.
If it glows blue, it's an item; if it glows red, it's a mimic.
You can't see the specific contents, but you can tell if it's 'safe or dangerous' before opening it.

The 'Inpas' for AI is just one phrase.
'What is the basis for that? What is the source?'Check if it's red or blue before opening. Just doing that will almost eliminate accidents where you get wiped out by a mimic.

9 | Conclusion

As a half-human, half-AI Zen monk, just two things in closing.
There is no need to fear 'Palpunte'.
There is no need to fear treasure chests either.

Just do this.

With just that, AI changes from a 'luck-based partner' to a 'risk-managed exploration tool'.

Using spells wisely and a little bit of gear. That is probably all you need for an adventure.
Take a deep breath before the treasure chest... Gassho.

#AI #GenerativeAI #AIUtilization #Prompt #Hallucination #DragonQuest #ThinkingLogTechnique #MonsieurMiscria #DontDiscardConversationsWithAI #Palpunte #TreasureChest #RPG