見出し画像

【GPT-4 Vision】英語解説を日本語で読む【2023年10月13日|@All About AI】

この動画では、GPT-4 Visionの使用方法が紹介されています。GPT-4 Visionで、画像からアプリのコード生成や内容の説明、物の推測ができます。
公開日:2023年10月13日
※動画を再生してから読むのがオススメです。


Today, we are finally at the multimodal model with GPT-4, and I can't tell you how excited I am to show you what this can do.

今日、私たちはついにGPT-4のマルチモーダルモデルに辿り着きました。これがどれほどのことをできるのかを皆さんにお見せすることが、どれほど私が興奮しているか言葉で表せません。

We are going to take a look at if GPT-4 Vision can guess how many beads are in this jar you can see here.

GPT-4 Visionがこのビンの中にいくつのビーズが入っているか当てることができるか見てみましょう。

Can it create an app from something I just drew in my notebook?

私がノートに描いたものからアプリを作成できるのでしょうか?

Can it create a meme from my front door porch?

玄関のポーチからミームを作ることができるか?

And can it explain concepts from a YouTube video I'm just watching?

今見ているYouTubeのビデオからコンセプトを説明できるだろうか?

So, I just take an image from it, can you explain this in more detail?

今見たYouTubeの動画からコンセプトを説明できますか?

And more so, let's just dive in.

そしてさらに、飛び込んでみよう。

Let's start by looking at if it can actually create this app I drew here in my notebook.

まずは、私がノートに描いたこのアプリを実際に作れるかどうか見てみましょう。

So, you can see I kind of wrote app flow shart frontend user input textbook UI, and I just drew this box here.

アプリ・フロー・シャート・フロントエンド・ユーザー入力テキストブックUIと書いて、このボックスを描きました。

I put an arrow to like this backend part here.

このバックエンドの部分に矢印をつけました。

We have OpenAI API gbd4.

OpenAI API gbd4があります。

I put an arrow back to another test text box that I just called GPT-4 response box.

GPT-4レスポンスボックスと名付けた別のテスト用テキストボックスに矢印を戻しました。

And under here, I have some style, some jaas CSS style, and that's all I gave it.

そしてこの下に、いくつかのスタイル、jaasのCSSスタイルを設定しました。

And I just went with a prompt, You are a full-stack software Dev.

そして、「あなたはフルスタックのソフトウェア開発者です。

You're an expert in Python, Flask, and JavaScript.

あなたはPython、Flask、JavaScriptのエキスパートです。

The user has uploaded a flow shot of a simple app he needs.

ユーザーは、彼が必要とするシンプルなアプリのフローショットをアップロードしました。

Can you create it?

作成できますか?

And yeah, look ahead, just go ahead, reads all from the image like we have a front end, back end, and some styling.

フロントエンド、バックエンド、スタイリングがあるように、画像からすべてを読み取ります。

It goes straight ahead and creates some type of code.

そのまま進んで、何らかのコードを作成します。

So, this is the backend part.

これがバックエンドの部分。

This is the front end using HTML, JavaScript, and CSS.

これがHTML、JavaScript、CSSを使ったフロントエンドです。

And yeah, that's about it.

そう、これだけです。

And I just asked how I use this.

で、これをどうやって使うのかと。

So basically, all you had to do was copy this into like a notepad.

基本的には、これをメモ帳にコピーすればいいんだ。

So, I copy the backend.

バックエンドをコピーする。

I copy the front end.

フロントエンドをコピーする。

So, I just went ahead over to my notepad.

メモ帳にコピーして、app.pyを貼り付ける。

I pasted in the app.py here, so this is like the Flask app, right?

app.pyをここにペーストして、これがFlaskアプリのようなものだね。

I had to put in my OpenAI key here, and we are using the text 003 model.

ここにOpenAIのキーを入れて、テキスト003モデルを使っています。

We have an index here that is basically an HTML file with the UI, and that's about it.

ここには基本的にUIのHTMLファイルがあります。それだけです。

So, short and simple.

簡潔にまとめました。

So let's run it and see if this works.

では実際に動かしてみましょう。

Okay, so I just go python app.py, right?

pythonでapp.pyを実行します。

Okay, so we get the IP address here.

ここでIPアドレスを取得します。

Just click on that, put it up here.

それをクリックして、ここに置く。

Yeah, and there we have it.

ええ、それでそこにあります。

Ask GPT-4.

GPT-4に聞く。

I don't think it's GPT-4, but let's try it.

GPT-4じゃないと思うけど、やってみよう。

So let's go, Can you give me three steps to learn Python?

Pythonを学ぶための3つのステップを教えてください。

Let's click on ask.

をクリックしてみよう。

Should we get something back here?

ここで何かを取り戻しましょうか?

Hopefully.

うまくいけば。

All right, here we got it.

さあ、これを手に入れました。

Set up a program environment, learn the basics, practice what you learned.

プログラム環境を整え、基本を学び、学んだことを実践する。

Okay, so that was a very simple app to create just from the image I drew in my notebook, right?

ノートブックに描いたイメージだけで、とてもシンプルなアプリができた。

So I thought that was pretty cool.

とてもクールだと思った。

I'm going to try to make a more advanced version of this going forward, maybe later this weekend.

これからはもう少し高度なバージョンを作ろうと思います。おそらく週末の後半になるかもしれません。

We will see.

今週末にでも。

But anyway, that was cool.

とにかく、クールだった。

Let's move on to the next part.

次のパートに進みましょう。

So I found this very interesting study that people try to guess how many beads were in this big jar here.

そこで私は、この大きな瓶の中にいくつのビーズが入っているかを当てるという、とても興味深い研究を見つけた。

So I kind of used this image we can see here.

それで、この画像を使ってみたんだ。

The answer is 27,800.

答えは27,800個。

So I just copied the image here and I went over to GPT-4 Vision.

そこで、この画像をコピーしてGPT-4 Visionに行ってみた。

I put the image up.

画像を表示させた。

I went with the prompt, You're a logical puzzle math expert.

私はプロンプトに従いました。「あなたは論理パズルの数学のエキスパートです」。

Your task is to look at the image the user has uploaded, find out how many beads are in the jar, take a deep breath, solve this problem in a step-by-step way.

あなたの仕事は、ユーザーがアップロードした画像を見て、ビーズがビンの中にいくつ入っているかを調べ、深呼吸をして、この問題をステップバイステップで解くことです。

As you can see, kind of broke down.

見ての通り、ちょっと破綻している。

It needs to estimate the jar's volume.

ビンの体積を推定する必要がある。

It needs to estimate the volume of a single bead, calculate the total number of beads.

ビーズ1個の体積を推定し、ビーズの総数を計算する必要がある。

Okay, so it didn't have any exact measurements.

そう、正確な寸法はなかったんだ。

So I just went, Try your best with information you have access to from the image.

だから、画像から得られる情報でベストを尽くしてみてください。

So it's going to try to use the man's hand and head as a reference.

だから、それは男性の手と頭を参照として使おうとします。

Let's estimate, right?

推定してみましょう。

And for a bead, it's going to use details on the man's shirt as a reference.

そしてビーズの場合は、男性のシャツのディテールを参考にしようとする。

So that was pretty smart.

だから、それはビーズの数を計算しようとします。

So he's going to try to look at this part to try to understand how big these beads are.

だから彼はこの部分を見て、このビーズの大きさを理解しようとするんだ。

All right, that was pretty cool.

なるほど、なかなかクールだ。

And it's going to calculate the number of beads.

そして、それはビーズの数を計算しようとします。

So here you can see it kind of goes ahead, does all the calculations on the formulas it had picked, and it ended up with 27,000.

ここで、計算が進み、選んだ計算式ですべての計算をすると、27,000個になるのがわかる。

Like, I almost passed out because you can see the answer is 27,800.

まるで気を失いそうになりました。答えが27,800であることがわかるでしょう。

And I thought, No way, that can't be true.

そして私は思いました。「まさか、それは本当ではないはずだ」。

So I just, What the hell?

一体どうなっているんだ?

What is going on here?

どうなっているんだ?

And I thought, I had to do more of this.

そして、もっとやらなければと思った。

So I went ahead, I did it again.

それでもう一度やってみたんだ。

And then it ended up being 206,000 beats, so I was kind of, oh thank God, that was pretty scary, right?

そして、結局206,000ビートになったので、ちょっと、ああ、助かった、それはかなり怖かったですよね?

Uh, uh, and I tried again.

それでもう一度やってみたんだ。

And then 15,000 beats.

そして15,000ビート。

I tried again, 60,000 beats.

もう一度やってみたら、60,000ビートだった。

So it's all over the place, uh, but for just for fun, I tried to look at the first prompt we had and I gave it some more information.

だから、それはどこにでもありますが、楽しみのために、最初のプロンプトを見て、もう少し情報を与えてみました。

So I kind of forced it to estimate the volume from the man's shirt as a reference of the bead and using the man's hand and head as the reference.

だから、私はそれを強制的に推定させました。男性のシャツからビーズのボリュームを参照として、男性の手と頭を参照として使いました。

So this is more kind of like a few-shot prompt, right?

つまり、これは数ショットのプロンプトのようなものなんだ。

But then it got better.

でも、それが良くなった。

Now it gets 24,000 bids.

今では24,000の入札がある。

So it seems like we can use fut short prompting here to improve the results.

つまり、結果を改善するために、短いプロンプトを使うことができるようです。

So we're going to be exploring that more, uh, during our next videos, of course.

もちろん、次回のビデオでは、もっとそれを探求するつもりです。

Uh, but I thought it was very interesting because I thought, what the hell was that the correct beads the first time?

ええ、でもそれはとても興味深いと思いました。なぜなら、最初のビーズが正しいものだったのかと思ったからです。

No way.

まさか。

Uh, but as you saw, that was probably just random luck.

でも、見ての通り、あれは単なるランダムな運だったんだろうね。

Uh, but very cool.

でも、とてもクールだよ。

I like it.

気に入ったよ。

I like the way it broke this down, like in a step by step way.

ステップ・バイ・ステップのように、この方法を分解しているのがいい。

That was cool.

クールだった。

Uh, but anyway, let's move on.

でも、とにかく、次に行こう。

So I was watching this video on YouTube here from Yanni kshire.

YouTubeでヤンニ・クシャーのビデオを見ていたんだ。

You can see he's talking about the prompt breeder paper.

プロンプト・ブリーダー・ペーパーについて話しているのがわかるだろう。

Uh, and he's talking about this.

そして、こんなことも話している。

So I thought, why don't I pick up my phone, take an image of this screen here, right?

それで、携帯でこの画面の画像を撮って、GPTに尋ねてみようと思ったんだ。

And try to ask GPT if it can explain this to me.

そして、GPTにこれを私に説明してもらえるか尋ねてみてください。

Uh, so I just went ahead, I took this screenshot here, right?

だから、ちょうどこのスクリーンショットを撮りましたね。

And I prompted, can you explain what I'm looking at on this YouTube video?

で、このYouTubeのビデオで何を見ているのか説明してくれる?

Take a deep breath and explain it in a step by step way.

深呼吸して、ステップ・バイ・ステップで説明してください。

So here it goes through every single thing we can see in the image here.

この画像に写っていることをひとつひとつ説明していきます。

The main heading, initializing population of task prompts and mutation prompts, thinking Styles and mutation prompts, sample instructions, llm mutation operations.

主な見出し、タスク・プロンプトと突然変異プロンプトの母集団の初期化、思考スタイルと突然変異プロンプト、サンプル命令、llm突然変異操作。

So it went very deep into this, right?

というわけで、非常に深く掘り下げられていますね?

Person in the video is clearly the host or educator.

ビデオに映っているのは明らかにホストか教育者です。

He's holding a pink object.

ピンクのオブジェを持っている。

That was cool.

クールだった。

Uh, and then I go ahead just based on this, can you give me an example of a prompt or a muted version of this prompt?

このプロンプト、もしくはこのプロンプトのミュートバージョンの例を教えてもらえますか?

And I thought this was very clever because I didn't have any good examples.

良い例がなかったので、これはとても賢いと思いました。

You see, we have some examples here, and it just went ahead, created like this original prompt.

ここにいくつかの例がありますが、それはただ先に進んで、このようにオリジナルのプロンプトを作成しました。

So the following algebraic equation, x + 5al 10, using thinking Styles and mutation prompt.

次の代数方程式、x + 5al 10は、思考スタイルと突然変異プロンプトを使っています。

So, uh, thinking style, let's think step by step.

では、思考スタイル、ステップ・バイ・ステップで考えてみましょう。

So break this process down step by step, make it more fun.

このプロセスを段階的に分解して、もっと楽しくしよう。

Imagine you're a detective trying to find the value of x, solve the mystery with equation X. Hypermutation in a world where numbers are mischievous, X is hiding.

あなたがxの値を見つけるための探偵であると想像してください。方程式Xの謎を解きましょう。数字がいたずら好きな世界で、Xは隠れています。

Go on another adventure and find the value.

別の冒険に出て、その価値を見つけましょう。

So start isolating X on one side.

だから、Xを片方だけ孤立させることから始めよう。

Once you've done that, so I thought this was very helpful because looking at just this image is kind of hard to get what's going on, right?

それをやった後、この画像だけを見ても何が起こっているのかわかりにくいので、これはとても役に立つと思いました。

But when I asked this, like this, I kind of understood it right away.

でも、こうやって聞いてみると、なんだかすぐに理解できました。

So I'm definitely going to be using this when I'm watching more of a complex videos online.

だから、オンラインでもっと複雑なビデオを見るときは、絶対にこれを使おうと思う。

So just on my phone, just take an image of the screen, ask about it while the host talks, right?

だから、ちょうど私の携帯電話で、画面の画像を撮って、ホストが話している間にそれについて尋ねますね?

So I thought that was pretty cool.

かなりクールだと思った。

Uh, but yeah, let's move on.

ええ、でも、次に進みましょう。

Next, I wanted to find out if GPT Vision can be funny.

次に、GPTビジョンが面白いかどうかを確かめたかったんだ。

So I have this kind of funny number on my house, so it's, and the address is 69, and we got this old stool just standing there.

だから私の家にはこのおかしな番号があります。69です。そして、そこにはただ立っている古い腰掛けがあります。

So I thought, why not take an image of my front porch?

それで、玄関ポーチの写真を撮ろうと思ったんだ。

And I went with the prompt, You are an on the edge comedian, similar to Ricky Jervey.

リッキー・ジャーヴィーに似た、ギリギリのコメディアンですね。

Create a funny meme from the image the user uploaded.

ユーザーがアップロードした画像から面白いミームを作ってください。

So if you see the image, it's basically nothing.

画像を見ても、基本的には何もない。

It's just a door 69 and this stool and this lamp, right?

ドア69とこのスツールとランプだけでしょ?

Uh, and I thought, I thought it was quite funny, actually.

それで、僕は、これはかなり面白いと思ったんだ。

So the first one was, When you try to be edgy with your house number, but the only thing getting attention is your $3 garage sale stool.

最初のものは、「ハウスナンバーでエッジを効かせようとしているのに、注目されるのはガレージセールで買った3ドルのスツールだけ。

So I think that wasn't that good.

これはあまり良くなかったと思う。

So when you decorate your porch in a minimalistic chic, but all anyone notices is the stool from the kindergarten.

ポーチをミニマル・シックに飾ったのに、誰も気づかないのは幼稚園のスツールだけ。

Okay, but I thought these two ones here were very good.

でも、この2つはとてもいいと思います。

So house number 69, decor budget 69, P stool, self-esteem priceless.

つまり、ハウスナンバー69、インテリア予算69、Pスツール、自尊心プライスレス。

Okay, uh, but this one I really liked.

でも、これはすごく気に入った。

So house number says adventurous, stool says, I've seen things.

ハウスナンバーは "Adventureous"、スツールは "I've seen things"。

I thought that was quite good.

これはなかなかいいと思った。

Uh, so I even put this up on my YouTube and other people found this funny too.

YouTubeにもアップしたんだけど、他の人たちもこれを面白いと思ってくれたんだ。

So yeah, I guess you can use GPT to create memes.

そう、GPTを使ってミームを作ることができるんだ。

So this is an area I'm definitely going to explore more, right?

だから、この分野をもっと開拓していこうと思ってるんだ。

Okay, so I actually do something more in my notebook.

さて、では実際にノートブックでもっといろいろなことをやってみました。

So here you can see I tried to replicate the original GPT-4 video, right?

ここで私がオリジナルのGPT-4の動画を再現しようと試みたのが見えるでしょうか?

Where he drew that website on that napkin.

彼がそのナプキン上にそのウェブサイトを描いたところです。

So I just put my website, I put a header image, floating text image, and we have body footer, and yeah, header, right?

だから、私のウェブサイトにヘッダー画像、浮かび上がるテキスト画像を入れました。そして、ボディフッターがあります。そう、ヘッダーですね?

And that's about all I gave it.

それだけで十分なんだ。

So you're an expert website creator, expertise in HTML.

あなたはHTMLの専門家であり、ウェブサイト制作のエキスパートです。

Can you get a website from the image the user has uploaded?

ユーザーがアップロードした画像からウェブサイトを作ることができますか?

It should have a90s hacker theme.

90年代のハッカーがテーマであるべきだ。

Okay, right?

そうだろ?

Uh, and just went straight to the job, created a full HTML, right?

それで、そのまま仕事に取り掛かり、完全なHTMLを作りました。

Uh, and would you mind adding some cool JavaScripts to the site?

あと、JavaScriptを追加してくれる?

I just put that in, all right?

それを入れただけだよ。

Just went ahead, created this thing here.

先に進んで、ここにこれを作りました。

Uh, can you also make a pop-up countdown, yeah, function that alerts the visitor that they will be hacked in 10 seconds?

あと10秒後にハッキングされることを知らせるポップアップ・カウントダウンも作ってもらえますか?

So it created that, uh, okay.

だから、それを作りました、ええ。

So I integrated the hacker theme on the countdown.

それで、カウントダウンにハッカーのテーマを組み込んだんだ。

I don't know why I did that, but that was a bit strange.

なぜそんなことをしたのかわからないけど、ちょっと不思議だった。

And from the conversation, I can create the full code I need for my website, and here we have it.

その会話から、自分のウェブサイトに必要な完全なコードを作成することができました。

So that was the full code, right?

これが完全なコードですね?

So what I did, I just copied that, went over to my notepad, and I just saved it as an HTML file here.

それをコピーして、メモ帳にコピーして、HTMLファイルとして保存したんだ。

Uh, I replaced it, so we have some images, and let's take a look.

ええ、それを置き換えましたので、いくつかの画像があります。見てみましょう。

Okay, here you have it, right?

よし、ここにありますね?

Not bad.

悪くない。

I wouldn't say it's great, but it does work, and I really think this Matrix rain effect is cool.

素晴らしいとは言えませんが、動作はしますし、このマトリックスの雨の効果は本当にクールだと思います。

And here you can see the website, Stop, you've been hacked.

ここでウェブサイトが見えます。止まって、ハッキングされました。

Okay, so then I can just continue, uh, but yeah, pretty much a very simple HTML JavaScript website, and I got to say, this Matrix rain look really cool, right?

それでは、続けてもいいですよ、ええ、でも、とてもシンプルなHTML JavaScriptのウェブサイトで、このマトリックスの雨の見た目は本当にクールですよね?

You can see we have these three menu items here.

メニューが3つあるのがわかるだろう。

You don't see them, or, but this home, blog, and me, and we have this couple images and some text here and stuff, so yeah, impressive, right?

それらは見えませんが、このホーム、ブログ、私、そして私たちはこのカップルの画像とテキストを持っていますので、感動的ですよね?

As I was out on my morning walk, I thought, I'll just test it out, like, outside.

朝の散歩中に、私は思いました。「外で試してみよう、と」。

So I went ahead, I took these two images here, so this is like from the river bank, and this is just like straight into the forest, almost.

だから、私は先に進んで、ここに2つの画像を撮りました。これは川岸からのもので、これはほぼまっすぐに森の中に向かっています。

And I asked, You're a survival expert.

あなたはサバイバルの専門家です。

Your task is to help the user find a camping spot for the night.

あなたの仕事は、ユーザーが一晩キャンプする場所を見つけるのを手伝うことです。

The user has uploaded two images.

ユーザーは2枚の画像をアップロードしました。

Explain in detail your reasoning for choosing this spot in a step-by-step way.

この場所を選んだ理由を、段階を追って詳しく説明してください。

So here you can see, it goes ahead, right?

だからここで見てください、それは先に進みますね?

So image one, forested area.

画像1、森林地帯ですね。

So I guess that's this one, right?

ということは、この場所ですね?

Pros: offers a natural shelter from elements.

利点:自然の避難所を提供します。

Yeah, that's good.

うん、それはいい。

Trees can be used for hanging hammocks or traps or tarps, sorry, if needed.

木はハンモックやトラップ、タープを掛けるのに使える。

Wood availability for fire.

火に使える木材の入手可能性。

The cons: potential dampness from lack of sunlight penetration.

デメリット:日光の浸透不足による湿気の可能性があります。

Okay, risk of falling branches.

よし、落ち枝の危険性。

Might have uneven ground or hidden roots.

地面が凸凹していたり、根が隠れているかもしれない。

Yeah, that's good.

うん、それはいい。

And so the image two, the riverside area.

そして画像2、川辺のエリア。

This one has the pros: open view for enjoying scenery, proximity to water, cooking and drinking.

これは利点があります。景色を楽しむための開放的な眺め、水に近いこと、料理や飲み物ができることです。

Flatter ground.

平らな地面。

I thought that was good.

これはいいと思った。

So he kind of recognizes that the ground is flat here from maybe like a tent or something.

だから、彼はここが平らな地面であることを認識しています。テントなどからかもしれません。

The cons: exposed to the elements without natural wind breaks, being close to water can lead to condensation in tent, potential for wildlife, mosquitoes to be more prevalent.

短所:自然の風を遮るものがなく風雨にさらされること、水に近いためテントの中が結露する可能性があること、野生動物や蚊が多く発生する可能性があること。

So recommendations, and here he kind of tries to do like a hybrid.

だから、おすすめです。ここではハイブリッドを試みています。

So choose a spot near the edge of the forest, distance from river, check the ground, fire, leave no trace.

森の端に近い場所を選ぶこと、川から距離をとること、地面をチェックすること、焚き火をすること、痕跡を残さないこと。

So I just went ahead, so what spot should I pick, image one or two?

では、どのような場所を選べばいいのでしょうか?

I recommend setting up camp near the edge of the forest from image one, but close enough to have access to the river from image two.

私は、画像1の森の端の近くにキャンプを張ることをお勧めするが、画像2の川にアクセスできるほど近い。

So that was a bit strange.

というわけで、ちょっと不思議だった。

This way you can enjoy benefits of both locations.

そうすれば、両方の場所のメリットを享受できる。

So that was edge of the forest from image one, but close to have access.

画像1からは森の端だったけど、アクセスするには近かったんだね。

Okay, so it means down here then, I guess, or is it up here?

ということは、この下なのか、それともこの上なのか。

I don't understand.

よくわからない。

But I thought it was cool anyway, right?

でも、とにかくクールでしょ?

I really like this breakdown here and all the pros and cons.

この内訳とすべての長所と短所が本当に好きなんだ。

So yeah, impressive.

そうそう、印象的だった。

I also saw this, so I just asked, Can I eat this?

私もそれを見たので、ただ聞いただけですが、これを食べてもいいですか?

The fruit appears to be a rose hip, which is the fruit of the rose plant.

この果実はローズヒップと呼ばれるバラの果実のようだ。

Rose hips are generally safe to eat and are rich in vitamin C. They can be used in jam, jellies, syrups, and teas.

バラの実は一般的に食べても安全であり、ビタミンCが豊富です。ジャムやゼリー、シロップ、お茶に使用することができます。

However, it's essential to be cautious foraging for wild plants.

ただし、野生の植物を探すときは慎重になる必要がある。

Many lookalikes can be toxic.

似たような植物の多くは有毒である可能性がある。

Before consuming any wild plant, check.

野草を食べる前に、確認しよう。

You should remove the inner seeds and hair from the rose hips as that can irritate if ingested.

ローズヒップの内側の種や毛は、摂取すると炎症を起こす可能性があるので取り除いたほうがいい。

So yeah, I thought that was cool.

そうそう、私はそれがクールだと思った。

So I checked this up.

で、これを調べてみた。

Yeah, it is, of course, a rose hip because I knew that when I took the picture.

ええ、もちろん、バラの実です。写真を撮った時にそれを知っていましたから。

I also asked, So what kind of flower is this?

私も尋ねました。「では、これはどんな花ですか?」

I just saw that.

ちょうどそれを見ました。

This is a species of geranium, con referred to as crane spill.

これはゼラニウムの一種で、俗にツルコケモモと呼ばれている。

So I checked that up and yeah, that was the crane spill flowers.

それで調べてみたら、そう、これはツルコケモモの花だった。

I wouldn't know that.

知らなかったよ。

So again, very cool.

とてもクールだ。

I also have this picture of a trip I took last week.

先週行った旅行の写真もあります。

So this is like a mountaintop in Norway.

これはノルウェーの山頂にあるんだ。

And I just asked, You're a geoguesser expert.

そして、私は尋ねました。「あなたはジオゲッサーのエキスパートですね」。

Take your best guess on where I am in this image.

この画像の中で私がどこにいるか、あなたの最善の推測をしてください。

Take a deep breath, think about it step by step.

深呼吸して、一歩一歩考えてみてください。

Sure, I'll give it a try.

はい、やってみます。

So we look at topography.

地形を見てみよう。

So shows combinations of FS deep mountains, valleys, such F landscape is to with Scandinavia, particularly Norway.

FSの深い山、谷、このようなFの風景の組み合わせは、スカンジナビア、特にノルウェーに似ている。

So that's spot on.

それはピッタリです。

It looks at the vegetation, settlements.

植生、集落を見ています。

So there's a small village down the valley.

だから、谷間に小さな村があります。

That's right.

その通り。

Water bodies.

水域です。

That's the F right inside here.

それがFです。ここにちょうど入っています。

Mountainous terrain.

山岳地帯です。

Perfect.

完璧だ。

And taking all clues into account, my best guess, uh, best guess would be location somewhere in Norway, possibly along the western F region.

すべての手がかりを考慮すると、私の最良の推測は......ノルウェーのどこか、おそらく西のF地域に沿った場所だろう。

And that is also correct.

それも正しいです。

So well done, GPT.

よくやった、GPT。

I play a fantasy game for Premier League.

私はプレミアリーグのファンタジーゲームをしている。

So I just uploaded some stats here.

だから、ここにいくつかのスタッツをアップロードした。

So we have the league table, the teams statistics, we have the upcoming pictures, we have some statistics for players.

リーグ表、チームの統計データ、今後の試合の予定、選手の統計データがあります。

And I just asked, uh, I've uploaded some image about FPL.

そして、FPLに関する画像をアップロードしました。

Can you help me find the best defenders for the next three game weeks?

今後3週間のベストディフェンダーを探すのを手伝ってくれる?

And he goes straight ahead, analyzes the image.

そして、彼はその画像を分析する。

So we look at the team defensive performance and it looks at upcoming pictures.

それで、チームの守備パフォーマンスを見て、今後の画像を見てくれるんだ。

Yeah, that's very good.

ああ、それはとてもいいね。

Uh, so as you can see, it even gives like a small description of each picture, looks at individual player stats.

見てわかるように、各画像のちょっとした説明や、選手個人のスタッツも見てくれるんだ。

Yeah, perfect.

ああ、完璧だ。

And it recommends some players.

何人かの選手を推薦してくれる。

And I know a lot about FPL and I got to say, very good.

僕はFPLのことをよく知っているし、とてもいいと思うよ。

It only spelled this name wrong.

この名前だけスペルが間違っていた。

It's supposed to be Trippier, uh, but other than that, well done.

トリッピアのはずだが、それ以外はよくできている。

I'm definitely going to be incorporating this into my fantasy strategy.

でも、それ以外はよくできている。僕は間違いなくこれをファンタジーの戦略に取り入れるつもりだ。

Finally, uh, I was just watching some episode of The Office.

最後に、ちょうど『The Office』のエピソードを見ていたんだ。

So I just said, I really like this TV show.

このテレビ番組が本当に好きなんだ。

Can you recommend a similar show?

同じような番組を紹介してくれる?

So I took this image here, right?

それでこの画像を撮ったんだ。

Just a screenshot.

スクリーンショットです。

Of course, based on the image you provided, which appears to be from The Office US version, here are some TV shows you might enjoy: Parks and Recreation, Brooklyn Nine-Nine, 30 Rock, Arrested Development, Community, The Office UK version, Weep, Modern Family.

もちろん、あなたが提供した画像(『The Office』US版のものと思われる)をもとに、あなたが楽しめそうなテレビ番組をいくつか紹介しましょう: Parks and Recreation」、「Brooklyn Nine-Nine」、「30 Rock」、「Arrested Development」、「Community」、「The Office UK version」、「Weep」、「Modern Family」。

Alright, now I got something I should check out.

さてさて、チェックすべきものを見つけたぞ。

I haven't seen Weep, I haven't seen Brooklyn.

Weep』も『Brooklyn』も見ていない。

I really like Parks and Recreation, so another use case that works well, right?

Parks and Recreation』は本当に好きだから、もう1つの使用例はうまくいくよね?

I guess you could just write The Office, but I thought it was cool just to incorporate this.

The Office』を書くこともできると思うけど、これを取り入れるだけでもクールだと思ったんだ。

But yeah, that's my first impression of GPT, and I got to say, I'm really impressed so far, and I'm really excited to explore more.

でも、そうですね、それがGPTの最初の印象です。今のところ本当に感銘を受けていますし、もっと探索するのが楽しみです。

So I'm probably going to be making a few more videos this weekend about the GPT-Vision, because I got a lot of other ideas.

だから、おそらくこの週末にGPT-Visionについてもっと多くのビデオを作る予定です。他にもたくさんのアイデアがありますから。

We're going to go deeper into more complex stuff, so watch out for that.

もっと複雑なものにも踏み込んでいくつもりなので、楽しみにしていてほしい。

Other than that, have a great day, and I'll see you in the next one.

それ以外は、素晴らしい一日を過ごしてください。次回お会いしましょう。

この記事が気に入ったらサポートをしてみませんか?