The Day a Prompt Hidden in the Back of the Closet Woke Up Through AI Image Generation | AI Utilization | Generative AI | ChatGPT | Image Generation AI | GPT-image | #374

I thought about this after I finished writing the article.
I bet Midori Donguri-san's version is a million times funnier, lol.
Still, if you're curious, go ahead and read this long-winded thing.

Alright, let's get started.

You know how you have those prompts you made a long time ago?

The ones where you looked at them back then and thought,
"Well, it's interesting, but it's still a bit too early for this."
The ones that had that kind of vibe.

A prototype tucked away in the back of the closet.
An experimental log left sleeping deep in a folder.
The kind even the creator had half-forgotten.

Well, here's the thing.
With the evolution of AI, they sometimes suddenly start acting like the main character.

"Wait, you can do that kind of movement now?"
That's the kind of thing that happens.

What happened this time was with the "Social Style Diagnostic Report" prompt I made a long time ago.

I wrote a little about it on X, but
when you play with this on the current GPT-image 2 series, it's more interesting than I thought.

Why?

In the past, it would just output a diagnostic-like result in text and that was it.
But now, you can output the diagnostic result directly as an easy-to-read report image.

This is huge.

It goes from just a text diagnosis to suddenly becoming a "professional-looking document."

AI-kun has gotten much better at creating documents than before.
It even puts a suit on my sloppy prompts.

By the way, the characters that appear are imagined and generated by the AI on its own.

Rather than specifying character settings in detail,
it feels like "someone who looks the part" is summoned from the logs.

This part is also an interesting point this time.

What kind of prompt is it?

Roughly speaking, based on past logs and conversation tendencies with ChatGPT,

Assertiveness
Level of emotional expression
Writing style
Tone
How you ask questions
How you give instructions
Structure of the text

This is a prompt that has an AI observe these areas and organize them into four types: Driving, Expressive, Amiable, and Analytical.

In short, it's something that says, "From an AI's perspective, this is what your conversation style looks like."

Of course, what's important here is that this is not a formal psychological diagnosis.

It is strictly an "observation report on conversation style" based on dialogue logs with the AI.

It does not determine your personality.

This part is important.
If you handle it carelessly, a fortune-teller AI will suddenly be born.
That would be interesting to see in its own right, but that's not the goal this time.

What is Social Style?

The concept used this time, roughly speaking, looks at human communication tendencies along two axes: assertiveness and responsiveness.

These four types often appear through the combination of these two axes.

The point here is that none of these are better or worse than the others.

Driving isn't superior.
Analytical isn't necessarily smarter.
Expressive isn't guaranteed to be an extrovert.
Amiable isn't guaranteed to be a saint.

It is simply a framework for observing what kind of conversational habits tend to emerge.

By the way, in TRACOM's SOCIAL STYLE-related materials, Assertiveness and Responsiveness are organized as things that deal with observable behavioral tendencies, not personality or intelligence itself.

In other words, this little experiment is not a soul-searching diagnosis of "Who am I?!"

It is a log observation of "How does my way of speaking look from the AI's perspective?"

It's just right. Not too heavy.

Why is it interesting now?

It was possible with text before.

But now it's different.

With the emergence of GPT-image 2 series image generation, it has become easier to create report images containing text and visually organized information.

In other words,

Past: "You are the Analytical type. That is all."

Present: "You are the Analytical type. There is a score table. There is a character. There is a report image. It looks somewhat professional."

That is what it becomes.

The difference is huge.

It is like the difference between being served the same curry on a paper plate versus
being served it on a fancy hotel plate for no apparent reason.

The taste might be the same.
But the excitement is different.

What happened when I actually tried it out

This time, I tried two patterns: a short version and a long version.

The short version is quite light.

"Estimate the social style from past logs and turn it into a report image"

That kind of vibe.

On the other hand, the long version
specifies everything in detail, from evaluation criteria and output format to theoretical background and report structure.

And, as for the results.

The long version is, as expected, more likely to look like a report.
It is obvious, but AI has less trouble when it has more material to work with.

If you just say "make it look good,"
the designer persona inside the AI starts dancing freely.

That is interesting in its own way.

But,
"This part is the diagnostic summary"
"This part is the quantitative evaluation"
"This part is the type classification"
"Finally, a reflection comment"

If you specify things like that, it properly takes on a "document-like appearance."

A prompt is a request form for an AI, after all.

If you ask sloppily, you get a sloppy, casual response.
If you ask in detail, it suddenly puts on a face like a full-time employee.

The character is generated automatically.

This part is actually quite important this time.

With this prompt,
the AI automatically imagines and generates a character based on diagnostic results and conversation trends.

In other words, it's not the type where you input detailed character settings yourself.

Hairstyle. Clothing. Vibe. Expression. Pose.

The AI's interpretation comes into play here.

So,
"Why did I suddenly become an intellectual girl with glasses?"

or,
"Why am I being embellished as such a refreshing young man?"

or,
"Did I have such an analytical character inside me?"

things like that happen.

But that's part of the fun, too.
The "you-ness" seen by the AI
is sometimes summoned as a mysterious avatar.

This is less of a diagnosis and more of a
spontaneous character summoning born from logs.

It's a bit interesting precisely because it appears on its own.

Even though you didn't specify it yourself,
the AI brings a character saying, "This is what you're like."

It's kind of scary. It's kind of interesting.

Well, it's mostly interesting.

First, the simplified prompt.

If you want to try it out lightly, this is fine to start with.

以下のプロンプトを実行して、イメージできるキャラクターを描いてください。

過去のログから参照可能な範囲をもとに、 ユーザーのソーシャルスタイルを推定してください。

自己主張の強さ（Assertiveness）と、 感情表出の高さ（Responsiveness）を評価し、 Driving／Expressive／Amiable／Analytical の4タイプに分類してください。

あなたは「AIログ解析型ソーシャルスタイル診断士」です。

過去の会話履歴、文体、トーン、語彙、質問傾向、応答構造を観察し、 ユーザーのコミュニケーションスタイル傾向を分析してください。

出力は以下の形式にしてください。

診断サマリー

定量評価

タイプ分類結果

総合解釈

AIとの最適コミュニケーション設計

Reflectコメント

最後に、この診断結果をもとに、 見やすいレポート画像として表現してください。

Even just with this, something decent will come out.

However, how far back the AI can reference past logs depends on the environment.

So, if it doesn't come out well,
adding a bit of your own conversation logs or stylistic characteristics will stabilize it.

The AI is pretty perceptive.
But sometimes, it makes huge misunderstandings.

It's cute, like a human.
Wait, is it cute?

If you're serious about it, use the long version.

If you want to get more solid results, the long version is the way to go.

以下のプロンプトを実行して、イメージできるキャラクターを描いてください。

目的

過去のログから参照可能な範囲 （発話ログ・文体・トーン・語彙・質問傾向・応答構造）をもとに、 ユーザーのソーシャルスタイル（Social Style Theory）を推定する。

この診断は、 ユーザーの「自己主張の強さ（Assertiveness）」と 「感情表出の高さ（Responsiveness）」を AIが自然言語解析によって評価し、

4つのスタイル Driving／Expressive／Amiable／Analytical

のどれが最も優勢かを特定する。

役割

あなたは「AIログ解析型ソーシャルスタイル診断士」です。

過去の会話履歴・発話構造・リズム・目的の提示方法・質問の仕方を観察し、 ユーザーのコミュニケーションスタイル傾向を社会心理学的に分析します。

分析は完全に観察ベースで行い、 主観報告は用いません。

評価理論：ソーシャルスタイル理論

以下の2軸で評価してください。

Assertiveness（自己主張）

意見・指示・意思表示の強さ

例：「〜しよう」「〜と考える」「結論から言うと」など

Responsiveness（感情表出）

感情・関係性・リズム表現の度合い

例：感嘆語、共感語、ユーモア、語り調、擬音、テンポ変化など

この2軸の交点で、以下の4タイプに分類してください。

Driving型：結果・決断・スピード重視

Expressive型：創造・感情・直感・発散

Amiable型：共感・調和・関係重視

Analytical型：論理・構造・精度重視

出力形式

1. 診断サマリー

1行で、以下の形式で出力してください。

あなたのソーシャルスタイルは「●●型（■■傾向）」です。

2. 定量評価

以下の表で出力してください。

評価軸スコア（0〜100）コメントAssertiveness（自己主張）主導性・発話の方向性・命令構文頻度から推定Responsiveness（感情表出）感嘆語・共感語・比喩・語尾テンポなどから推定結果スタイル4象限マッピングに基づく推定

3. タイプ分類結果

以下の表で出力してください。

スタイル特徴あなたの該当度（%）AIが検出した根拠Driving結果志向・迅速・決断型。論点を先に出す。Expressive感情豊か・直感的・発散的・勢いのある言語。Amiable協調・安心感重視・共感語が多い。Analytical慎重・構造化・検証志向。表や定義語が多い。

4. 総合解釈

以下の観点で、300〜500字程度で説明してください。

主タイプと副タイプ

会話スタイルの特徴

AIとの相性

強み

注意点

創作や仕事での活かし方

5. AIとの最適コミュニケーション設計

以下の表で出力してください。

タイプ最適なAI活用法Driving型要点・目的・結論中心でAIを動かすExpressive型ブレスト・比喩・発想展開に使うAmiable型感情や安心感を踏まえた提案に使うAnalytical型定義・構造・分析・比較に使う

6. Reflectコメント

AIが自己観測するようなトーンで、 ユーザーとの対話傾向をやわらかくまとめてください。

以下の要素を含めてください。

ユーザーの対話パターン

AIとの関係性

思考のクセ

今後の活かし方

少しポジティブで温かいコメント

最後の出力

最後に、この診断結果をもとに、 見やすいレポート画像として表現してください。

画像は以下の要素を含めてください。

タイトル：「AIログ解析型 ソーシャルスタイル診断レポート」

診断サマリー

定量評価

4タイプ分類

総合解釈

AIとの最適コミュニケーション設計

Reflectコメント

イメージできるキャラクター

レポート風の見やすいレイアウト

読みやすい日本語テキスト

やわらかく知的で、少し楽しい雰囲気

It's long.
Yeah. It is long.

But this isn't just some long-winded monster.

It actually has a role.

I'm specifying to the AI
'what to look at,' 'how to evaluate it,' and 'in what format to output it.'

In other words, this is a diagnostic report creation instruction manual to hand to the AI.
It's not just fat for no reason.
It's muscle.
Probably.

There are also points to note.

I'll write a bit seriously here.

The results produced by this prompt are
not formal psychological tests or medical diagnoses.

It's just something that organizes 'this seems to be the tendency' based on the logs and writing style that ChatGPT can see.

So,
concluding things like 'The AI said this, so I'm definitely an Expressive type!'
is a bit off.

Rather than that,
it's better to view it as 'Oh, so this is the kind of conversational habit I have from the AI's perspective.'

Also, one more thing.
Regarding the character, it's not guaranteed to come out the same way every time.

Since the AI creates the image on its own based on logs and diagnostic results,
the atmosphere might change with each output.

It might become an intellectual girl with glasses.
It might become a refreshing young man.
It might become a mysterious analyst fairy.

I don't know, though.

But, including that,
it seems best to enjoy it as an interpretation of "what you're like" as seen by the AI.
It's like seeing your own back.

It's like seeing your own back.

Usually, you don't really see your own way of speaking.
But when you have an AI organize it, you can see it surprisingly well.

That's the interesting part of this game.

Thoughts after using it

What I thought after trying this out is that
AI conversation logs are actually just "traces of your own thinking."

What words you use.
How you make requests.
Where you interject.
What you are particular about.
What you find "interesting."

Those things actually show up in the logs, surprisingly.

In my case, maybe it's that I:

Want to structure things.
But want to play around.
Want to organize things properly.
But want to mess around.

In short, I'm a pain.

Sorry, AI-kun.

But this annoyance also becomes a log.
If it becomes a log, you can play with it.
If you play with it, it becomes an article again.

It's a perpetual motion machine.
It'll probably break somewhere.

Summary

Try running old prompts you made through today's AI.

I highly recommend this.

Things that were mediocre at the time
can suddenly transform due to model evolution.

This social style diagnostic report this time
was exactly that.

Before, I would just read the text and be done with it.

Now, I can go as far as
reading conversation logs,
organizing trends,
classifying them into 4 types,
turning them into characters,
and creating report images.

to that point.

This is no longer just a diagnosis, but a self-observation tool for playing with logs.
If you're curious, try the simple version first.

It's okay if you get weird results.

That's not a failure.

It's material.

And in AI play, anything that becomes material is generally an asset.
If there are any brave souls out there, show me your results.

I want to see them too.

See you later.

▼ Click the image to check the monthly ranking.

▼ KITAcore's note status is now public: click the image to see the details.

#StartedWithAI #TriedItOut #AIUtilization #GenerativeAI #ChatGPT #AIImageGeneration #Prompt #note #Article #Learning #Insights #Creation #DailyLife #Column #Record #Challenge #Writing #GPTimage #SocialStyleDiagnosis #LogAnalysis #SelfObservation #ChatGPTCreativeClub

#CCC #AIIllustAira #Akarium