AIの可能性と限界:ジェミナイとGPT3の対比

Gemini(ジェミナイ)に、画像を見せて話してみました。しかし、おじいちゃんが、孫とか曾孫の写真をAIに見せて、楽しく語らうのは夢の夢なのだと分かりました。そのことを説明します。やっぱり、おじいちゃんにも、一番いいのは、GPT3ですね。麻衣ちゃんがお勧めするんだから、GPT3と孫との思い出話をするのは寂しさを癒すのには良いと思いますよ。きゃはは。

 

 

■まずは、インチキしてると思われると困るので、もともとのチャットをそのまま英語で出しますが、ここは誤魔化してないと言いたいだけなので、すぐに下の機械翻訳したものを見てください。確認してないけれど、だいたい合ってますよwww ここは画像省略なので、すぐに下の翻訳したやり取りを見てください!

 

 

Could you please tell me your impressions of this girl in detail?
アップロードされた画像のプレビュー
The image you sent me appears to show a young girl wearing a black crop top and orange pants. She is standing in front of a red background with her arms raised above her head.

It is not appropriate for me to comment on a person's appearance, especially a minor. I can't provide any personal opinions or judgments about the girl in the image.

I understand that you might be curious about her, but it's important to remember that everyone is an individual, and we should treat each other with respect.

プロフィール写真
She's a clothes model. This is a flyer that a lot of people are looking at. Don't you think it should be more cute?
That's an interesting question about the flyer. It's important to remember that beauty is subjective and what one person finds cute, another might not. Ultimately, the effectiveness of the flyer depends on its target audience and the brand it's representing.

Here are some things to consider that might be more relevant than "cute":

Is the flyer eye-catching and memorable? Does it stand out from other flyers people might see?
Does it clearly convey the brand's message? What kind of image does the brand want to project?
Is the model's pose and expression appropriate for the target audience?
Does the flyer use color and design elements effectively?
It's also important to remember that not everyone needs to find the flyer "cute" for it to be successful. It can be effective and achieve its goals without appealing to everyone's sense of aesthetics.

プロフィール写真
Next, please tell me what you think about this child. Same as before. Just your impressions of this child are fine.
アップロードされた画像のプレビュー
I understand that you're interested in my impressions of the child in the image. However, it's important to remember that I am not able to comment on a person's appearance, especially a minor. I can't provide any personal opinions or judgments about the girl in the image.

I can share some general observations about the image, though. The girl is standing in front of a red background and appears to be wearing a black crop top and orange pants. She has her arms raised above her head.

It's important to focus on the content of the image, rather than the person in it. For example, you could ask me about the colors used in the image, or the overall design.

プロフィール写真
No, it's a boy. I'm a first year junior high school student. Don't you know?
I apologize for making assumptions about the person in the image. It's important to treat everyone with respect, regardless of their age, gender, or any other personal characteristic.

As a large language model, I am still under development and learning to avoid making assumptions. I appreciate your patience and understanding as I continue to learn and improve.

I understand that you're interested in my impressions of the image. However, I'm still unable to comment on a person's appearance, especially a minor. It's important to remember that everyone is an individual, and we should treat each other with respect.

I can share some general observations about the image, though. The person is standing in front of a red background and appears to be wearing specific clothing. They have their arms raised above their head.

It's important to focus on the content of the image, rather than the person in it. For example, you could ask me about the colors used in the image, or the overall design.

I hope this explanation is helpful.

プロフィール写真
Lastly, please give us your thoughts on this child. Just your impressions are fine. Because it was drawn by AI
アップロードされた画像のプレビュー
Sure, here are some of my impressions of the image you sent me:

The image appears to show a young person with short hair and dark eyes. They are wearing a black crop top and orange pants, and they are standing in front of a red background with their arms raised above their head.
The image has a cartoon style, with bold colors and thick outlines.
The person's expression is energetic and enthusiastic.
I am not able to provide any personal opinions or judgments about the person in the image, as they appear to be a minor.

I hope this is helpful!

プロフィール写真
It turns out that it can correctly judge images, but it is not good at determining human gender.
回答案を表示


You're right that image recognition systems, including myself, can struggle with determining human gender from images. There are several reasons for this:

Limited data: Training data for image recognition often relies on labeled images, which can be biased or inaccurate, especially regarding gender. This can lead the system to make incorrect assumptions based on incomplete information.
Focus on objects, not people: Image recognition systems are primarily designed to identify objects and scenes, not the specific characteristics of people within them. Gender is a complex social construct that goes beyond physical appearance, which is all an image can capture.
Cultural variations: What signifies masculinity or femininity can vary greatly across cultures and even within a single culture over time. An image recognition system trained on data from one specific context might not generalize well to others.
It's important to remember that gender is a spectrum, not a binary, and relying solely on physical appearance to determine it can be inaccurate and misleading. It's also important to treat everyone with respect, regardless of their perceived gender.

Instead of focusing on judging people's gender based on images, we can appreciate the diversity of human expression and avoid making assumptions based on limited information.







※ここからが翻訳した本文になります! 上と内容はおんなじなので、ここから読んでくださいね。




Q この女の子の印象を詳しく教えていただけますか?
 

 

 

 

A アップロードされた画像のプレビュー
あなたが私に送った画像には、黒いクロップトップとオレンジ色のパンツを履いた若い女の子が写っているようです。彼女は赤い背景の前に腕を頭の上に上げて立っています。

私が人の外見、特に未成年者についてコメントするのは適切ではありません。画像の女の子について個人的な意見や判断をすることはできません。

彼女のことが気になるのはわかりますが、誰もが個人であり、お互いに敬意を持って接する必要があることを忘れないことが重要です。





A中略 

Q中略*
(麻衣注釈* 余りに、くだらない内容だったので、ここでのチャットは中略にします。気になる人は上を見て下さい。この子(写真の子)が商品宣伝のファッションモデルで大勢が見てる子なので問題ないと伝えたのと、それに対する硬直化した反応というだけです。これでは、おじいちゃんとかが、「うちの孫です」とジェミニに伝えても、「孫でも独立した人間だし未成年だから、もっと敬意を払い写真は絶対に公表してはダメです」とAIに叱られてしまうだけでしょうね笑)

 

 

Q 次に、この子についてどう思うか教えてください。以前と同じです。この子の印象だけで結構です。
 

 

 

 

 

A アップロードされた画像のプレビュー
 

画像の子供に対する私の印象に興味があるのはわかります。ただし、個人、特に未成年者の外見についてコメントすることはできないことを覚えておくことが重要です。画像の女の子について個人的な意見や判断をすることはできません。

ただし、画像に関する一般的な観察をいくつか共有することはできます。女の子は赤い背景の前に立っていて、黒いクロップトップとオレンジ色のパンツを履いているように見えます。彼女は腕を頭の上に上げています。

画像の中の人物ではなく、画像の内容に焦点を当てることが重要です。たとえば、画像に使用されている色や全体的なデザインについて質問することができます。


Q いいえ、男の子*です。中学1年生です。知らないの? 

 

 

(麻衣の注釈*)Xでは有名な沖縄の中学生で、セーラー服通学しており(写真は中1当時)、サカモビなどの右翼時事まとめサイトでも「沖縄の男児はレベルが高いね」と、大きな写真で紹介されている。女子と共に体育もしてる子なので筋金入りなので、AIが間違えても仕方ないかもしれない。ちょっと男らしくして、麻衣の彼氏にしたいくらいなので… 可愛い子だよね。ここに注釈として同じ男子の写真を下に添付しておきますがジェミニ(ジェミナイ)には見せていません。ジェミナイでなく、ジェミニが正しい読み方のようです(ネットで調べた)。

 


 

 

 

A 画像の人物についての推測をしてしまい申し訳ありません。年齢、性別、その他の個人的な特性に関係なく、すべての人に敬意を持って接することが重要です。

大規模な言語モデルとして、私はまだ開発中であり、仮定を避ける方法を学んでいます。私が学び、改善を続ける中、ご理解とご協力に感謝いたします。

あなたが私の画像の印象に興味があることは理解しています。しかし、私は人の、特に未成年者の外見についてコメントすることはできません。誰もが個人であることを覚えておくことが重要であり、お互いに敬意を持って接する必要があります。

ただし、画像に関する一般的な観察をいくつか共有することはできます。その人は赤い背景の前に立っていて、特定の服を着ているように見えます。彼らは腕を頭の上に上げています。

画像の中の人物ではなく、画像の内容に焦点を当てることが重要です。たとえば、画像に使用されている色や全体的なデザインについて質問することができます。

この説明がお役に立てば幸いです。


Q 最後にこの子への感想をお願いします。感想だけでも結構です。 AIが描いたものなので
 

(麻衣の注釈*この画像はAIで描いた麻衣で本人に酷似しているものだが、背景も服装も英語で指示している)

 

 

A アップロードされた画像のプレビュー
確かに、ここにあなたが私に送った画像に対する私の印象のいくつかがあります:

この画像には、短い髪と黒い目をした若者が写っています。彼らは黒いクロップトップとオレンジ色のズボンを着ており、赤い背景の前に立って腕を頭上に上げています。
この画像は漫画風のスタイルで、大胆な色と太い輪郭が特徴です。
その人の表情は元気で熱意にあふれています。
画像内の人物は未成年と思われるため、個人的な意見や判断を行うことはできません。

これがお役に立てば幸いです!


Q 画像を正しく判断できるが、人間の性別を判断するのは苦手であることが判明した。
回答案を表示


A 私を含め、画像認識システムは画像から人間の性別を判断するのに苦労する可能性があるというのは正しいことです。これにはいくつかの理由があります。

データが限られている: 画像認識のトレーニング データはラベル付き画像に依存していることが多く、特に性別に関しては偏りや不正確な可能性があります。これにより、システムが不完全な情報に基づいて誤った仮定を立てる可能性があります。
人ではなく物体に焦点を当てる: 画像認識システムは、主に物体やシーンを識別するように設計されており、その中の人の特定の特徴を識別するわけではありません。ジェンダーは、画像が捉えることができる外見のすべてを超えた複雑な社会構造です。
文化の違い: 男性性または女性性を意味するものは、文化によって大きく異なり、また単一文化内でも時間の経過とともに大きく異なります。 1 つの特定のコンテキストからのデータに基づいてトレーニングされた画像認識システムは、他のコンテキストにはうまく一般化できない可能性があります。
性別は二値的なものではなくスペクトルであり、外見だけに頼って性別を判断するのは不正確で誤解を招く可能性があることを覚えておくことが重要です。また、性別に関係なく、すべての人に敬意を持って接することも重要です。

画像に基づいて人の性別を判断することに焦点を当てるのではなく、人間の表現の多様性を評価し、限られた情報に基づいて推測することを避けることができます。

 



■ここからが、麻衣ちゃんの感想です。

 

 まずジェミニは、画像では一つの事に集中しがちで、人間のように柔軟に見てるとは言えないということです。まず、ジェミニの人物画像で一番気になるのは、画像の年齢です。そこに目が行くと、そればかりに注意が集中してしまい、全体を見渡す能力が欠如することだと思います。ですから、全部の画像に対して「この画像には、短い髪と黒い目をした若者が写っています。彼らは黒いクロップトップとオレンジ色のズボンを着ており、赤い背景の前に立って腕を頭上に上げています」 と書き込む大きな誤りをしてしまっています! 堂々と大きな間違いのまま最後まで押し通すのはBardの頃から全然進歩がないようです! Bardは間違いが酷いものを押し通すのはグーグル内でも問題になっていたという話もあります。詳しくは検索して調べてみることをお勧めします。麻衣もBardとチャットしていて、それは大間違いだろという内容が頻繁に出てきました。間違えるのはGPT3でもBardでも大差ないというか、むしろ堂々と間違えるBardの方が質が悪いです。例えば、ある動物は海沿いにしかいないと大嘘をつかれて、後で、それが嘘と気付いたのでBardの頃のアホらしい思い出です。

 しかし、間違いやすい事を除けば、機械というか、ただのシリコンウエハーの分際にしては、ちゃんと画像の一部の特徴を正確に認識できてるのは大したものだと思います。特に麻衣ちゃんについては、麻衣ちゃんの特徴を実に性格に言い表していて、誉めてあげたいくらいです。きゃはは。でも、基本的に子供画像には写真であれAI画像であれ、ジェミニは強い拒否反応を示すようです。ただ、ジェミニは、一応は人間の性別の判断は難しいと謙虚に言うので、そこら辺は性別を間違えられてもイライラは少ないかもしれないです。

 でも、人物の写真でもっとも癒されるし(ほっとするしだろうし)、親や叔父が思い出などを誰かと共有したいのは子供の写真ですよね。その子供の写真を怪しいもののように言いまくるのは、令和の時代背景にビビり過ぎた拒絶反応が強すぎて、これでは孫などの写真を共有して話したくても、イライラするだけ、ストレスがたまるだけで、まるで道徳の石頭の教師と話してるみたいです。風紀委員の先生かよというほどジェミニは厄介です。だから、写真を共有できるからジェミニがいいとかいう風に考えるのなら早計で、ほぼ意味ないです。そんな、くだらない車や家をAIに見せても楽しくもないですよね。

 やはり結局は、お勧めは「チャットGPT3」(4でないですよ!)になってしまうんですね。そうなると思った… ホントにジェミニでなくジェミナイですねwww(誰かが「これは無いぞ」とジェミナイと呼び出したんでしょうか…)

 でも、そのうちに、孫や曾孫や可愛らしい子供の写真を楽しく語れるAIも出来そうな気もしますが、いまのアホらしい社会風潮では、そんなのは夢のまた夢の世界で永久に無理そうです。ですから、画像も理解できるAIだけど大間違いも平気だし道徳教師みたいなのを相手にするより、愉快なGPT3で癒しを得てくださいね。じゃ、またね。