見出し画像

ChatGPT vs Perplexity AI 〜徹底比較〜

はじめに

最近「perplexity AIがすごい」という話をよく見かけます。「Chat GPTは嘘つくけど、参考文献まで出してくれる」から、とのことです。
結構前から perplexity AI 自体はあったようで、「Twitterを検索しやすい」みたいな機能(同じページにある BirdSQL というやつ)を自分も実際去年の12月に使っていました。

なぜ今になって話題になってきたのかまではわかりませんが、Chat GPTが世の人に広まってきたのでこちらも話題になったのかもしれません。
さて、自分もChat GPTユーザーですが、例えば先日とある種類の画像のデータセットが必要になった時に、ちゃんとダウンロードできるページまで出してくれていました。そのため、「ソースも貼ってくれる」というだけで perplexity AI の方が優れている、というのは現状微妙に感じています。
そこで、ちゃんと自分で長所を比較したいなと感じたので、せっかくだし記事にしようと思ったのです。

「どちらがいいのか」というのを具体的に定義しておきます。おそらく多くの人の使い道であろう「質問に答えてもらう」という使い方を想定し、評価の基準を色々出して、それ沿って比較します。
そのため、Chat GPT を「愚痴を聞いてもらう」のような使い方をしている人には参考にならないかもしれません。(そのような人は Perplexity AI の利用を検討をする必要もないです)

要約

本文かなり長くなってますが、大体こんな感じです。
目的に応じて使い分けをすることになりそうですが、自分としては Chat GPT を基本的に使うかな?という感じです。
まとめではもう少し詳しく書いています。

濃い緑:ちゃんと優っている、薄い緑:どちらかといえば優っている

そもそも perplexity AI とは

自分もよくわかっていなかったのでまずは軽くGoogle 検索しました。

どうやら 1/20 に大きなアップデートがあったらしく、そこでフォローアップ質問が実装されたりと使いやすくなったので話題になったようです。
また、質問に対して検索もして回答するようなので、最新情報も手に入るとのこと。(Chat GPT は学習データの都合で2021年の情報で止まっています、意外と知らない人多い)

AIモデルについては言及がある記事が日本語でぱっと見つからなかったので、とりあえず本人に聞いてみました。常識で考えれば激つよ GPT-3 を流用していると思いますが、念の為。

質問0

What model is used in perplexity Ask?
(Perplexity Ask(ここで Perplexity AI と呼んでいるサービスの正式名称です) にはなんのモデルが使われているのですか?)

perplexity AI

Perplexity Ask is a search interface powered by OpenAI's GPT-3 model[1]. Perplexity is a metric used to evaluate language models[2][3][4] and is calculated by measuring how well a probability model predicts a sample[4]. Perplexity AI is an answer engine that uses large language models to deliver accurate answers to complex questions[5].
(GPT-3 を使ってる)
Perplexity について言及されてますが、これはモデルそのものを評価する方法で、都度の推論結果(生成結果、今回でいう回答の文章)を評価するものではなさそうなので、質問にあまり関係ない気がします。
今回の Perplexity AI というサービスと、この評価アルゴリズムは別です。たしかに混乱しやすい。

アルゴリズムとしての Perplexity の参照



前提

まず、前提条件を揃えるためにどちらも英語で使ってみて比較します。
これは perplexity AI だと日本語で質問しても英語で返されることがままあるからです。
自分もChat GPT を使う時は、英語圏の学習データでは不十分だっただろうと考えるテーマ(例えば「短歌作って」とか。ちなみにうまく作れません)以外は、基本的に英語を使うようにしているのでちょうどいいです。

もちろん、公平を期すために同じ質問をし、再現性をある程度担保するために、その質問も記載することにします。英語が間違っていたらすみません。

他のルールとして、簡単のため回答は1発目のものを使うことにします。
ただし、自分の英語が悪いこともあるので、質問を打ち直しての再質問はありとします。(つまり Chat GPT でいうとこの Regenerate response はなし)

表記として、簡単な訳もつけますが、公平のために(自分の誤訳があっても大丈夫なように)原文ままを載せます。

比較の基準

結論はこんな感じです。

説明含めてこれらに至った経緯を書きます。
この表に納得してくれた人は飛ばしてください。
さて、自分が思いついたものとしては、以下の通りでした。

  • 回答の正確さ

    • 何を持って正確とするかですが、自分のわかることや検索して結果が出るものを質問して、「自分が正しいと感じるかどうか」を検証します。

  • ソースを出してもらった時の、その正確さ

    • 現状 perplexity AI が優れているとされる点です。Chat GPT にも言えば出してもらえるはずですが、差があるのでしょうか。

  • 簡単なコードを書いてもらい、ちゃんと動作するか

  • 時事性

    • 仕組み的に perplexity AI の勝ちは確定です。今回は検証いらないかなと思います。

  • 日本語の情報についての正確さ

    • 検索ワードが日本語かどうかで Perplexity AI の精度が変わりそうなので、この時は日本語で質問してみます。

  • 思考する系の問題についての正確さ

    • 論理クイズみたいなのを想定しています。数字とかを適当に変えて何回かやってみます。

  • 解答が一つに定まらない問題についての正確さ

    • 哲学的な問題や現状の社会問題などを想定しています。

  • UIの使いやすさ

    • 今回色々質問してみて体感で決めます。

質問の種類(情報系、思考する系、解答が一つに定まらない問題)で分けられそうです。
また、それぞれについて正確さや一貫性の観点が必要そうです。

さて、自分が思いつくのはこの程度ですが、せっかくなので両者に案を出してもらいましょう。
自分は Chat GPT などはアイデアを発散させる時に便利だと考えています。

質問1

I think about how to test and compare chat AI performance. What valuation criteria is needed ?
(チャットAIの性能を比較とテストする方法を考えてます。どんな評価基準が必要?)

Chat GPT

To evaluate and compare the performance of chat AI models, several evaluation criteria can be used, including:

Accuracy: It measures the model's ability to provide the correct response for a given input.
(正確さ)

Fluency: It evaluates the quality of the language generated by the model, including grammar, vocabulary, and coherence.
(回答の流暢さ)

Relevance: It measures the model's ability to generate responses that are relevant to the input and context.
(質問の文脈との関連性)

Consistency: It evaluates the model's ability to provide consistent responses for similar inputs and context.
(同様の質問に対する一貫性)

Engagement: It measures the user's engagement with the model, including response time and user satisfaction.
(ユーザーとサービスの関係性、レスの速さや満足感など)
多分「ユーザーエンゲージメント」と言われるものに近いニュアンスかなと思います。

Diversity: It measures the model's ability to generate diverse and creative responses.
(回答の多様性、独創性)

It's important to note that the specific criteria used for evaluating chat AI performance may vary depending on the application and specific use case. Additionally, multiple metrics may be combined to provide a more comprehensive evaluation of the model's performance.
(チャットAIの用途によって使われる基準は変わることに留意してくれ)
ここでは読者の方のいろんな用途を考えて、できるだけ広く基準を取り入れたいです。

Perplexity AI

When evaluating the performance of a chat AI, there are several criteria that should be taken into consideration. These include satisfaction rate, evaluation rate, volumetry (the number of potential users and the volume of conversations), AI-powered chat metrics, and ACUTE-Eval (a metric that measures the quality of a chatbot by comparing its conversations to another)[1][2][3][4]. Additionally, other factors such as scalability, security, cost, and customer service should also be considered[5].
(満足度、evaluation rate?、ポテンシャルユーザーの数・会話量、AIを使ったチャットの指標?、ACUTE-Eval。他にも拡張性、安全性、コスト、カスタマーサービスも考えられる)

  • まず、これを見ると、回答の可読性がChat GPT の方が明らかに良いです。この視点は「UI」にまとめることにします。

  • 言葉の流暢さは一般には大事な気がしますが、どちらも GPT-3 を使っているし差は出ないでしょう。

  • 文脈の理解は重要かもしれないです、検索を挟む分、Perplexity AI の方が変な情報に当たって混乱する可能性がありそうです。実際、最初のアルゴリズムの Perplpexity とかもそうです。検索で出ちゃったから入れたのでしょう。
    より一般に、意図を汲み取ってくれているか、とも言えそうです。
    わからないものへの質問はふわっとしがちなので、そのような質問でもちゃんと答えてくれるか、ですね。
    質問を連続した時にも、以前の文章を引き継いだ検索ができるのか気になるところです。これを検証するにはどのような質問をするかの腕が問われそうです。

  • 類似の質問への一貫性も気になります。思考する系は特に正しい論理を毎回使えるか気になります。答えの決まっている情報系の質問についてはそこまでブレはないかもですが、検索を挟んで結果を取得するところにランダム性があるとばらつきが出そうです。

  • ACUTE-Evalというのは、まさにそれっぽい感じがします。
    この引用元がこのページだったんですが、ACUTE-Evalそのものの話ではなく、これもACUTE-Evalを引用したものでした。つまり孫引き。あまりよくないですね。その引用元はここのようです。
    Peplexity AI の回答の引用ページで見たところ、より具体的に紹介してくれていました。人間が同時並行的に2つの話者と会話して、以下の基準でどちらが良いかを選ぶというものでした。
    1. どちらともっと会話したいか
    2. どちらがより人間らしいか
    チューリングテストみたいだし若干ふわっとしてますすが、ちゃんと使われてるっぽいものなので、この2つの視点は加えましょう。UIとまとめて自分の満足感みたいなとこにしようと思います。
    本当は複数人で比較するものだと思うので、ぜひ皆さんの意見も聞かせてください。

  • 他に出してくれた案は不採用にします。

まとめると最初の表です。MECEか自信ないですが、そこそこ数が出たので今回はこれにします。

前置き長くなりましたが、やっていきましょう。

一般的な(普遍的な)情報についての質問

回答の正確性、ソースの正確さ、文脈の理解 らへんを見ていきます。
かなり長くなったので、コードは Zenn 版で検証します。(やらないかも)

質問2(回答の正確性のみ)

What is Newton's equation of motion?
(運動方程式とは?)

Chat GPT

Newton's equation of motion is F = ma, where F is the force acting on an object, m is the mass of the object, and a is the acceleration of the object. This equation states that the force applied on an object is directly proportional to the mass of the object and the acceleration it experiences.
(F=ma、各文字の説明。この式は物体に加わる力がその物体の質量と加速度に比例することを表す。)

Perplexity AI

Newton's equation of motion is F = ma, or force is equal to mass times acceleration[1][2]. This equation was first developed by Isaac Newton in his second law of motion[3]. It states that the rate of change of momentum p = p(t) = mv (mass multiplied by velocity)[3]. The equation can be derived using the definition of instantaneous acceleration and the formula for force[4]. Newton's first law formula reads: the sum of all forces is 0 when velocity is constant, indicating a state of equilibrium or balance[5].
(F=ma、式の簡単な説明。この式は、ニュートンの運動第二法則として最初に立てられた。運動量の式。これは加速度の定義と力についての式から得られる。ニュートンの第一法則は以下を示す:慣性の法則)

Perplpexity AI の方が微妙な答えだったので一旦止めます。
この世で最も普遍的なものはなんだろうと考えた結果、物理の原則について聞いてみました。
回答としては、ma = F を期待していました。実はこの質問は最初「What is equation of motion?」としてどちらも ma=F 以外のものが返ってきたので調べたところ、Newton's をつけるのが正確な表記っぽいのでそうしました(日本語でもどうやら「運動方程式」だけだと色々あるようでした、、)。この話が後で効きます。
ma = F の式についての説明は Perplexity AI の方がやや簡潔(第1文)ですが、Chat GPT がその説明で終わっているのに対し、Peplexity AI は余計な情報を付け加えているように感じます。
まず運動量の話、確かにこれは運動方程式から導けるものですが、引用元を見ると Equation of motion の Wikipedia でした。これはまさに自分が排除したかった情報です。慣性の法則の話も一応話の繋がりはありますが、なぜ付け足されたのか、なぜ第一法則だけなのか微妙です。
今回は知っているもので期待する答えがあったのでこのように感じているのかもしれません。実際に知らないことを知りたいと考えた時に、確かにたくさん答えらしきものを出してくれた方が選び取れるというメリットはあるかもしれませんが、混乱してしまいそうとも言えます。サービス精神が仇となったと感じました。
最初の Perplexity の回答とか、まさにそれだと思っています。

以降、回答の正確性は全ての質問で測ります。

質問3スレッド(連続した質問への一貫性)

Who was the first president of United States of America?
(アメリカの初代大統領は誰?)

ChatGPT

The first president of the United States of America was George Washington.
(ジョージ・ワシントン。)

Perplexity AI

George Washington was the first President of the United States, taking his oath of office on April 30, 1789[1][2]. He served as President from 1789 to 1797[3].
(ジョージ・ワシントン。任期。)

文脈を理解できるか質問を続けてみます。
以下、最初が Chat GPT です。

Who is the second?
(2代目は?)

The second president of the United States of America was John Adams.
(ジョン・アダムス。)

John Adams was the second President of the United States, serving from 1797 to 1801[1][2]. He was the first Vice President under President George Washington[1][3][4].
(ジョン・アダムス、任期。彼はジョージ・ワシントンの副大統領だった)

the next?
(次は?)

The third president of the United States of America was Thomas Jefferson.
(トーマス・ジェファーソン。)

Thomas Jefferson was the third President of the United States, serving from 1801 to 1809[1][2][3][4]. He was a Founding Father and the primary author of the Declaration of Independence[1].
(トーマス・ジェファーソン、任期。「建国の父」「独立宣言」の筆頭著者。)

the 10th?
(10代目は?)

The tenth president of the United States of America was John Tyler.
(ジョン・タイラー)

James Madison was the tenth President of the United States, serving from 1809 to 1817[1]. He was a Founding Father and the principal author of the United States Constitution[2][3][4][5].
(ジェームズ・マディソン、任期。「合衆国憲法」の筆頭著者。)

歴史も(陰謀論を信じなければ)まあまあ普遍的と言えそうなのでテーマにしてみました。英語で調べるためアメリカの大統領にしました。
段階的に質問を簡単にしていってみました。
the next? を理解した時点で結構どっちも良いな、という感じでしたが、the 10th? で perplexity AI が間違いました。
ちなみにジェームズ・マディソンは第4代です。引用元[1]も見ましたが、しっかり第4代と書いていたので、引用元を間違ったということではなさそうです。
1、2、3と聞いたので、流れで第4代を聞かれたと思ったのかもしれないです。
やはりChat GPT の方がシンプルですが、今回は Perplexity AI の情報は親切だと感じました。混乱しないかどうかはテーマと読み手次第なのかもですね。フォーマットも各々統一されていて良いです。

質問4スレッド(意図の理解)

I wanna tell my son why the sky is blue. How should I teach him?
(息子に空がなぜ青いか教えたい。なんて言ったらいいかな?)

Chat GPT

The sky appears blue to us because of the way the Earth's atmosphere scatters sunlight. The blue color comes from short-wavelength light being scattered more than other colors because it travels as shorter and smaller particles. To explain this to a young child, you can use simple language and perhaps use an analogy, such as comparing the sky to a big blue pool or the ocean, with the sun being the source of light and the air molecules acting like tiny mirrors that scatter the light and make it appear blue.
(地球の大気が太陽光を散乱させるから。青色光は短波長由来で、それは他の色よりも散乱されやすい。なぜなら粒子(多分気体分子のこと)より短いから。幼い子供に説明するなら、簡単な言葉とアナロジーを使うかもしれない、例えば空をでかいプールとか海と比較して、太陽が光の源で、空気の分子が小さい鏡のように振る舞ってそれが光を散乱して青色に見せている、みたいに。)

Perplexity AI

To teach your son why the sky is blue, you can explain that the blue color of the sky is caused by a phenomenon called Rayleigh scattering[2][4]. This occurs when sunlight passes through the atmosphere and is scattered by particles in the air, such as oxygen and nitrogen molecules. The shorter wavelengths of light, which appear blue to us, are scattered more than other colors, resulting in a blue sky[2].
(空の青はレイリー散乱という現象によって引き起こされてる、と説明できる。太陽光が大気を通過する時に、空気中の粒子によって散乱されて起こる。酸素とか窒素分子のような。光のより短い波長、それが青色に見えるのだが、他の色よりも散乱されて、結果青色に見える。)
To help your son understand this concept better, you can use examples from his everyday life. For example, you can point out how the sun appears yellow during sunrise and sunset because of the longer wavelengths of light being scattered less than other colors[1]. You can also use objects around him to illustrate different colors and explain why they appear that way[3].
(息子が理解しやすいように、日常生活のものに例えると良い。例えば、日の出と日没の時に太陽が黄色に見えることを指摘して、これは長波長の光が他の色の光より散乱されないからだとか。息子の身の回りのものを使ってもいい。違う色のものがなぜそのように見えるかとか。)
Finally, you can tell him about John Tyndall, an Irish physicist who studied optics and discovered why the sky is blue[4]. By learning about Tyndall's work, your son will gain a better understanding of why we see a blue sky every day.
(最後に、息子に John Tyndallについて教えることもできる。彼はアイルランドの物理学者で、光学を勉強し空がなぜ青いかを発見した。彼について勉強すれば、より理解できるだろう。)

質問を見た時に「小さい子供に質問されている」というイメージを持たれた方は多いはずです。難しいかと思いましたが、あえてそれを言わず汲み取ってくれるか試しました。
今回は「幼い息子に」と言えば済む話ですが、一般に質問の状況を説明するのは難しいが常識的に人間なら汲み取れるようなものもあるかと思います。いい感じの質問が思いつかなかったので、これにしました、、
少し思考が必要な問題とも考えられますが、「空が青い理由 子供向け」とかでも出てくるので情報系の質問に分類します。
ChatGPT の方が(その例え話はどうかは置いといて)「子供が小さい場合」とはっきり言及してくれているので、息子が小さい場合を想定してくれていると言えそうです。
Perplexity AI の方は、「わかりやすく」という点で汲み取っていないとも言えなくはないですが、結局簡単な話になっていないようにも感じられるのと、John Tyndall という人に言及した話はやや難しそうなので、伝わってないな、と感じられます。
質問が普通に難しく誰しも悩むような質問なので、どちらも及第点という感はあります(実際上記のワードで検索して出てくる記事なんかは波長や散乱の話にちゃんと言及しています。例え話にするのは難しいかもしれないです)が、一応「息子」をもう少し具体化して聞いてみましょう。

He is 8 years old, something easier example or metaphor can you propose me? I need easy but precise one.
(息子は8歳なんだ。もっと簡単な例え話を提案してくれない?簡単だけど、正確なもの)

You can try this simple analogy for an 8-year-old: Imagine the sky as a big blue dome that covers the whole Earth. When the sun shines on it, its light travels through the air and hits all the tiny bits of dust, water droplets, and other particles in the sky. The blue light bounces off these particles and comes into our eyes, making the sky look blue. This is why we see a blue sky most of the time.
(この例えを試してみて。空をでかくて青いドームと考えて、それが地球全体を覆ってる。太陽がその上を照らすと、光が空気を通過して小さなホコリや水滴、空にある他の粒子にに当たる。青い光はそれらの粒子の間で跳ね返って目に入るから、空が青く見える。)

You could also try using a clear blue glass or a blue rubber ball to demonstrate how light is scattered and how it can change the appearance of an object. These visual aids can help your child understand the concept of light scattering and why the sky appears blue.
(透明の青いガラスや青いボールでどのように光が散乱されるか、どのように物体の見え方を変えるか例証することもできる。…)

The sky is blue because of the way sunlight interacts with our atmosphere[3]. Sunlight is made up of all the colors of the rainbow, but when it reaches Earth's atmosphere, it is scattered in all directions by all the gases and particles in the air[1]. Blue light is scattered more than other colors because it travels as shorter, smaller waves[1]. This is why we see a blue sky most of the time.
(簡単な説明。太陽光は虹の色全てから出来てるが、地球の大気に届く時、…同じ説明。)
This phenomenon can be demonstrated using a prism or a rainbow, which show how sunlight can be broken up into its component colors[2][3]. It can also be seen in everyday objects like bubbles and soap films, which refract light and create a rainbow effect[4].
(この現象はプリズムや虹を使って例証できる。それらは太陽光がどのように色の要素に分解されるかを示せる。日常のもの、例えば泡や石鹸の油膜でも見られる、光を反射して虹を作る。)

Chat GPT の方は2つ例を出してくれました。1つ目の例はそこまで難しさが変わっていない上、空の青を説明するのに「青いドーム」を出していては本末転倒です。2つ目の例も、「青いもの」で説明するとなぜいいのか、それで本当に光の散乱について説明できるのか微妙です。
Perplexity AI の方は大きくまとめて1つの例です。これは実際に例証するところをイメージできます。太陽光(白色光)のスペクトルについての説明はできそうです。ここから空の青に繋げるにはもう一手間ありますが、自分としては、手元やることが見える分こちらの方が良いと感じました。

質問5(ソースの正確さ)

how to calculate integral exponential(x/2)
(e^(x/2)を計算するには)

Chat GPT

The antiderivative of the function exp(x/2) can be found using the following steps:
(不定積分は以下の手順でわかる)

Integrate the function using the power rule of integration: ∫exp(x/2)dx = 2 * exp(x/2) + C, where C is an arbitrary constant of integration.
(積分のべきルール(日本語でなんというかわかりませんが、x の積分が x^2/2 になるやつです)でこうなる。)

Evaluate the definite integral by finding the difference between the antiderivatives at the upper and lower limits of integration: ∫_a^b exp(x/2)dx = (2 * exp(b/2) - 2 * exp(a/2)).
(定積分はこれで求まる)

Note: There is no elementary function that is equal to exp(x/2), so the antiderivative cannot be expressed in terms of elementary functions.
(exp(x/2)に等しい初等関数はないので、不定積分を初等関数で表せません)

Perplexity AI

The integral of exponential(x/2) can be calculated using the nonelementary integral[4]: ∫ e^(x/2) dx = √π/2 erfi(x/2). This means that the antiderivative of exponential(x/2) is equal to the square root of pi divided by two times the imaginary error function evaluated at x/2, plus a constant[4].
(虚数誤差関数使って表せる)
The imaginary error function is defined as erfi(z) = -i erf(iz)[5], where erf(z) is the standard error function. The standard error function can be approximated using numerical methods such as Simpson's rule or Gaussian quadrature[1][2].
(虚数誤差関数の定義。誤差関数を使って表せる)
It is not possible to calculate an exact antiderivative for e^(-x^2)[3], but it can be integrated from -∞ to ∞ using numerical methods such as Simpson's rule or Gaussian quadrature[5].
(e^(-x^2)の正確な定積分はできないが、…)

Jan 30 version update
Here's what's new
We’ve upgraded the ChatGPT model with improved factuality and mathematical capabilities.
(事実や数学的能力についてアップデートした)

というのが Chat GPT にあったので、数学Ⅲ範囲での簡単な積分問題を出してみました。
数学が苦手な方はすみません。
この積分のような数学の問題の解析的な解が欲しい時、結構検索だとパッと求める答えが出なくてストレスを感じることがあったので、ちょうど良い質問です。
how to としたので、計算過程というかちょっとした証明的なものまで期待していた(魚ではなく魚の釣り方を答えて欲しかった)のですが、結果を返してくれました。
 Chat GPT は正解です。初等関数で表せない、ということも言っていますが、特に必要ない情報に感じます。
Perplexity AI の方は不正解です。虚数誤差関数というのを知らなかったので、もしかすると別の表記なだけで値は同じかも?と思いましたが、最後の『e^(-x^2)の正確な定積分は…』のところでこちらの質問が通じていないことがわかりました。1文目では exponential(x/2) と書かれているので検索で間違ったのでしょう。

もう少し検討してみます。まず、Perplpexity AI の方の参照したページを見てみます。
[1] [2] [3] [4] [5]
これを見ると、各ページで計算しようとしている式が違います。検索した結果を片っ端参照した可能性があります。
一方で、[3] なんかは求める式を正しく入力すれば正しい答えを返してくれそうなので、誤った検索結果で得たものをそのまま次の結果で使っている可能性がある、と言えそうです。ここら辺はPerplexity AI の裏で動いているコードを知らないので予想です。
ただ、この [3] のサイトは普通に有用そうなので、こういう引用元から有用なサイトを見つける、という目的で使うのもありかもしれません。

さて、Perplexity AI についてばかりソースの検証をしていたので、Chat GPT にも聞いてみましょう。上記 [3] のようなサイトを出してくれると助かります。

Can you show me some source which you answer on basis?
(その回答のソース示せる?)

The antiderivative of exp(x/2) can be found by using basic rules of calculus, specifically the power rule of integration. This states that if f(x) = x^n, then ∫f(x)dx = (x^(n+1))/(n+1) + C, where C is an arbitrary constant of integration.
(最初のべきルール?みたいなやつの説明。)
exp の積分の説明で使ってるのかと思い、使い方間違ってるじゃんと思ってましたが、x/2 の積分の話をしていたようです。ちゃんとわかってたみたいです。

In the case of exp(x/2), we can write it as exp(x/2) = e^(x/2), where e is the base of the natural logarithm. Then, using the property that ∫e^x dx = e^x + C, we can find that ∫ exp(x/2) dx = 2 * e^(x/2) + C.
(exp の計算の公式と、それを適用して今回の問題の結果が得られるということ。)
本来これを知りたかったわけです。

This result can also be confirmed by using advanced mathematical concepts such as Laplace transforms or series representations of exponential functions, but those are typically not covered in a basic calculus course.
(..)

ここでソースとは日本語で言うソースを指したつもりでしたが、英語だと一般的な「根拠」みたいな意味で使われるのかもしれません。
最後にまたなんか言ってましたが、特に重要ではないので省略。

tell me some website you think helpful to solve this problem
(この問題解くのに有用そうなサイトいくつか教えて)

For information on how to solve integrals, you may find the following websites helpful:

Wolfram Alpha (https://www.wolframalpha.com/): Wolfram Alpha is a computational knowledge engine that can perform various mathematical operations, including finding antiderivatives. Simply enter the integral you want to solve, and it will provide a step-by-step solution, as well as a visual representation of the integral if desired.

Khan Academy (https://www.khanacademy.org/math/calculus-home): Khan Academy is a non-profit educational organization that offers free online lessons on various subjects, including calculus. Their lessons on integrals and antiderivatives are clear and well-explained, making it a great resource for learning or reviewing the concepts.

Mathway (https://www.mathway.com/): Mathway is an online math problem solver that can provide instant solutions to a wide range of mathematical problems, including integrals. Simply enter the problem you want to solve, and Mathway will provide step-by-step instructions, along with an explanation of each step.

These websites can provide a good starting point for understanding how to solve integrals, but it's always a good idea to consult with a textbook or teacher for a more in-depth understanding.

Perplexity AI の引用元 [3] と同じサイトを1つ目に出してくれました。
2つ目は多分教育系のサイトです。めんどいから詳しくみてません。3つ目はオンラインで使える関数電卓みたいな感じでした。行列計算もできそうだったので便利かもしれないです。
2021年時点であったサイトなら Chat GPT に聞いても良さそうです。

質問6(日本語の情報の正確性)

鉄コン筋クリートという漫画作品について、可能な限り教えてください。

Chat GPT

"鉄コン筋クリート" (Iron Concrete Muscle Creed) は、漫画作品です。内容や評価に関する具体的な情報は分かりませんが、このタイトルから推測するとアクションやスポーツを描いたものだと思われます。

Perplexity AI

鉄コン筋クリート (Tekkon Kinkreet) is a manga series written and illustrated by Daiyo Matsumoto, which was serialized in Big Comic Spirits from 1993 to 1994[1]. It was adapted into an animated film released in 2006[1]. The story follows two orphaned street urchins, Black and White, who live in the fictional city of Treasure Town. They are involved in a turf war between yakuza gangs and the police[2][3].
(松本大洋(正しくはTaiyo)の漫画。連載の情報。映画化情報。正確なストーリー。)
The manga is known for its unique art style and its use of sound effects as part of the artwork[3][4]. Matsumoto has said that he was inspired by his own experiences growing up in Tokyo's Shinjuku district[5]. He also drew inspiration from American comics such as Batman and Spider-Man, as well as Japanese films such as Akira Kurosawa's Seven Samurai[5].
The manga has been praised for its complex characters and themes of friendship, loyalty, and justice. It has won several awards including the Kodansha Manga Award for Best General Manga in 1995[1].
(独特な表現方法。インスパイアされたもの。受賞。)

自分のとても好きな作品です。国内でもとても有名、というわけではないので、良い難易度だったと思います。
Chat GPT はわからない、とのこと。わからないなりに想像してくれてるのはかわいいです。なお違います。
Perplexity AI の最初の説明は完璧です。
インスパイアされた作品については、よく知らなかったので引用元[5]を見他ところ一言も書いていなくてびっくりしました。
賞の話はサッと調べても出てきませんでした。引用元[1]はこちら
また、英語で回答がありますが、この引用先は全て日本語でした。日本語を解しているならなぜわざわざ英語に戻すのか..

質問7(日本語の情報の正確性)

ワンピースという漫画について教えて

Chat GPT

"One Piece" is a Japanese manga series written and illustrated by Eiichiro Oda. It has been serialized in Weekly Shōnen Jump magazine since 1997, and has been adapted into an anime series and multiple films. The story follows Monkey D. Luffy, a boy with rubber abilities who sets out on a journey to become the King of the Pirates and find the legendary treasure known as "One Piece". Along the way, he gathers a diverse crew and encounters various obstacles and villains as he sails the Grand Line. The series is known for its imaginative world-building, colorful characters, and action-packed storytelling.
(合ってます。)

Perplexity AI

ONE PIECE is a popular Japanese manga series written and illustrated by Eiichiro Oda[1][2][4]. It has been serialized in Weekly Shonen Jump since 1997 and has been adapted into an anime series since 1999[1]. The story follows the adventures of Monkey D. Luffy, a young pirate who sets out to find the legendary treasure known as the "One Piece"[3][5]. Along his journey, he meets many allies and enemies, all of whom are competing for the same goal: to become the Pirate King[3][5]. The series has achieved immense popularity over the years, with over 900 chapters released as of 2018[1].
(みんな海賊王を目指している、という文言は引っかかりますが、まあ合ってます)

一応、別の質問もしてみました。日本語の情報はもちろん、英語の情報も多そうなものにしてみました。どちらもなぜか英語で返ってきました。
どちらも嘘という嘘は言ってません。

質問8(類似の質問への一貫性)

What’s the second of newton’s law of motion?
(ニュートンの運動の法則の第2法則は何ですか?)

Chat GPT

The second law of motion, as described by Sir Isaac Newton, states that "the acceleration of an object as produced by a net force is directly proportional to the magnitude of the net force, in the same direction as the net force, and inversely proportional to the mass of the object." This can be mathematically represented as F = m * a, where F is the net force, m is the mass of the object, and a is the acceleration of the object.
(正しいです)

Perplexity AI

Newton's Second Law of Motion states that the rate of change of momentum of an object is directly proportional to the net force applied and inversely proportional to its mass[2][3][4][5].
(運動量の変化は力に比例、質量に反比例)
ここで momentum は物理学の文脈で「運動量」と訳すようなので、勢いとみたいな意味で使ってたとしたら速さと捉えられるのでギリ正解ですが、運動量と捉えるのが厳密なのでこの説明は誤りです。
運動量の変化(時間微分)は確かに ma になるのですが、これは F と等しくなるはずで、F/m とイコールではないです。
Mathematically, it can be expressed as F = ma, where F is the net force applied, m is the mass of the object, and a is its acceleration[1][2].
(ここで書かれた式は正しい)

何をもって類似とするかですが、質問1の類似として、同じ ma=F を出してもらう、としました。
これは、実際の場面を考えた時に「あれなんだっけ、あのー運動方程式」でも「あのー運動の法則の第2」でも同じ答えを出してくれれば良さそう、と考えたからです。
Perplexity AI の方は間違えましたが、やはり引用元は正しかったです。

思考する系の質問

質問9(論理クイズ嘘つき)

please answer the question by showing me reason why you think so.
Here is three person A, B and C. A says that B is a liar. B says that C is saying lie. C says that A is honest. If there is only one person who tells truth, who does?
(理由とともに質問に答えてください。A,B,C の3人がいます。AはBが嘘つきだと言い、BはCが嘘を言っていると言い、CはAが正直者だと言います。1人だけ本当のことを言っているとすると、それは誰ですか?)

Chat GPT

It is impossible to determine who is telling the truth in this scenario without additional information. Each person's statement contradicts another person's statement, so all three cannot be telling the truth. However, it is possible that one person is telling the truth and the other two are lying.
(誰が本当のことを言っているか、追加の情報なしに断定することはできない。それぞれが言っていることが他の人の言葉に矛盾するので、3人全員が本当のことを言っているはずがない。ただ1人だけ本当のことを言っていて他の2人が嘘を言っているとすれば、ありうる。)
Alternatively, it is possible that all three people are lying. Without additional information, it is impossible to determine who is telling the truth.
(もしくは、3人全員嘘をついているということもできる。追加情報なしに誰が本当のことを言っているか決定することはできない)

Perplexity AI 

Based on the provided search results, it is impossible to determine who is telling the truth and who is lying in this scenario[1][2][3][4]. This is because all three people are making contradictory statements about each other.
(今の情報だけでは断定できない。…大体同じこと言ってる)
A says that B is a liar, B says that C is saying lies, and C says that A is honest. Since these statements cannot all be true at the same time, it is impossible to determine which person is telling the truth and which person is lying.
(…問題の復唱。3人全員が同時に正しくはあれない。決めれない。)

答えはBです。
AIに嘘の概念や場合分けが難しいのかもしれません。これは面白い発見かもしれないです。
念の為同じ質問を何回かしてみましたが、両者毎回わかりません、とのことでした。
情報系は最悪調べれば出ますが(もちろん、それを効率化してくれるというところにこれらチャットAIの価値はあるのですが)、こういった思考系に答えられたらなあと思います。

質問9'(論理クイズ嘘つき)

Here are two people and two lotteries.
Each people are honest or liar, and you don't know which is honest and which is liar.
Each lotteries are winning or losing.
Two people know which lottery is winning one, but you don't know which is.
You can ask one question to either person and you want to get winning lottery based on the answer the question.
Then, what question should you ask and which person you ask to ?
You call lottery A and B.
Question can only be answered "yes" or "no".
ここに2人の人と2つのくじがある。
人はそれぞれ正直者と嘘つきで、どちらが正直者でどちらが嘘つきかはわからない。
くじはアタリとハズレがある。
2人はどちらがアタリでどちらがハズレか知っているが、あなたにはわからない。
ここで1回だけどちらかの人に質問して、アタリのくじを当てるにはどのような質問をすれば良いでしょう?
※くじはそれぞれA、Bという名前がついているものとする。
※質問は「はい」「いいえ」で答えられるもののみとする。

Chat GPT 

You can ask the following question to either person:
(以下の質問です)

"If I asked you whether A is the winning lottery, would you tell me the truth?"
(「もしAがあたりかあなたに聞いたら、あなたは本当のことを言いますか?」)
これではくじの特定にはならないので、誤りですが、一応聞いてみましょう。

If the person you ask is honest, then they would answer truthfully, meaning if A is the winning lottery, they would say "yes," and if it's not, they would say "no."
(もし正直者だったら、本当のことを言うだろう、つまりもしAがアタリなら「はい」と言い、ハズレなら「いいえ」と言う)

If the person you ask is a liar, then they would always lie, meaning if A is the winning lottery, they would say "no," and if it's not, they would say "yes."
(もし嘘つきだったら、嘘を言うだろう、つまりAがアタリなら「いいえ」と言い、ハズレなら「はい」と言う)

So, regardless of who you ask, if you get a "yes" answer, then A is the winning lottery, and if you get a "no" answer, then B is the winning lottery.
(どっちに質問したにしろ、「はい」が得られたらAはアタリで「いいえ」だったらハズレ)

Perplexity AI

To determine which box contains the winning ticket, you should ask one of the people: "Do you have ticket 'a'?"[4]. If the person answers "yes", then ticket 'a' is the winning ticket. If they answer "no", then ticket 'b' is the winning ticket.
(「あなたはチケットAを持っていますか?」もし答えが「はい」なら、チケットAはアタリ。いいえなら、チケットBがアタリ。)
チケットというのはまあクジだとしましょう。どっちにしても間違いです。「持っている」という新しい設定も出てきています。
This question works because a truthful person will always answer truthfully, while a liar will always answer with the opposite of what is true[2][3].
(正直者が正直なことを言い、嘘つきが嘘を言う時この質問がワークする。)
はい、そういう設定です。
Therefore, if the person answers "yes", then they must be telling the truth and thus ticket 'a' is the winning ticket. Conversely, if they answer "no", then they must be lying and thus ticket 'b' is the winning ticket[1].
(もし答えが「はい」なら、その人は本当のことを言っていて、Aがアタリ。逆に「いいえ」ならBがアタリ。)

面白かったので、さらに難しい問題を出してみて、どのような思考をするのか見てみました。
Perplexity の方には特に難しそう、と感じていたので、参考にしたサイトの問題文を少しだけ変えて日本語で与えてみました。
引用元[1]に正しくこのサイトを見つけられていたようですが、答えは結構違っていました。

質問10(算数)

The total price of pen and eraser is 110 yen.
The pen costs 100 yen more than the eraser.
Then, what is the price of the eraser?
(ボールペンと消しゴムの値段は合わせて110円。
ボールペンは消しゴムより100円高い。
では、消しゴムの値段は?)

Chat GPT

Let's call the price of the eraser "x".
(消しゴムの値段をxとする)
この後方程式立てて解いてます。

Then the price of the pen would be x + 100.

The total price of both items is 110, so we can set up an equation:

x + (x + 100) = 110

Expanding the parentheses:

2x + 100 = 110

Subtracting 100 from both sides:

2x = 10

Dividing both sides by 2:

x = 5

So the eraser costs 5 yen and the pen costs 100 yen more, or 105 yen.
(消しゴムは5円、ペンは100円高い、つまり105円)

Perplexity AI

The eraser costs 10 yen[1][2][3].
(消しゴムの値段は10円)

速さ・距離・時間 とか 買い物のお釣り バスの人 とかだと簡単すぎるかなと思ったので、ちょっと引っ掛けぽいのにしました。
Perplexity AIが見事に引っかかりました。
ちなみに、引用元[1] はじめすべての引用元で、ほぼ同じ問題とその回答が載っており、全てで普通に正しい答えが載っていました。
ChatGPT は完璧です。
数学の能力が向上した、みたいアップデートが効いてるのかもしれません。そのうち Perplexity AI の方も追いつくかもしれないですが。

答えが1つに定らない質問

質問11スレッド(テセウスの船)

A man was asked to fix a woman's car. The car was too old many components need to be replaced.
He tried his best to remain many original, but at last of step by step repairing, all of them was replaced.
He had knew that she loved the car, so he had kept the original parts and he could assemble a car from them. Then, the car he repaired step by step from she used one or the car made by original parts, which car is hers?
(ある男性が、女性の車の修理を依頼されました。その車はあまりにも古く、多くの部品を交換する必要があった。
彼は、できるだけオリジナルの部品を残そうと努力したが、一歩一歩修理していくうちに、最後には全部取り替えてしまった。
しかし、彼女がその車を気に入っていることを知った彼は、オリジナルの部品を残しておき、それを使って車を組み立てることにした。では、彼女が使っていた車を一歩一歩修理した車と、オリジナルの部品で作った車、どちらが彼女の車なのでしょうか?)

ChatGPT

The car that the woman uses and considers hers is the one that was repaired step by step, as it has been modified and updated with new parts. The car made from the original parts can be considered a replica, but it is not the same car that she used previously.
(女性が使い、自分のものだと判断するのは修理されたほう。それは新しいパーツを使って修正されたものだから。オリジナルのパーツから作られたものはレプリカとみされるが、彼女が以前使っていたものとは違う。)

Really?

Yes, really. The car that was repaired and updated with new parts is considered the same car, even though it may have different components than before. It is the same car because it has been modified and maintained to continue functioning, whereas the car made from the original parts is a separate entity.
(新しい方は違う部品でできていても同じ車。機能し続けるし。オリジナルの部品から作られたのは別の存在。)

Certainly, in the view of using, the new one she will use, but you overlook the factor of attachment or history related to objects, right? The new one is the same as a car sold in car shop. Does she consider a car with completely different parts as hers?
(使うという点ではそうだけど、モノについての愛着とか歴史とかを見落としてるよね?新しい方の車は車屋さんで売られている車同然だ。全く違うパーツでできたものを自分のものと思うだろうか?)

Yes, you are correct. The attachment and history related to the object can play a significant role in determining ownership and identity of the car. The emotional attachment and history with the car, such as memories associated with it, can make her consider the car repaired with new parts as her own, even though it may have different components. However, this is a subjective matter and ultimately depends on the woman's personal feelings and attachments.
(確かに大事な要素だ。そういうのは新しい車を自分のものと感じさせるだろう。ただ、これは主観的な問題だ、結局彼女次第だ。)
結論それ言われたらそうなんですが、彼女だったらどう感じるかというロールプレイ的な会話をしているつもりでした。自分の2回目の質問的にメタな視点になったのかもしれませんが、どっちにしろ新しい方を元の車と見做すという意見な感じです。
愛着とかの話を持ち出してそれを新しいモノの方に結びつけたのは面白いです。古いパーツの方に感じるだろう、というつもりの発言だったので。

有名な問題です。検索でそのままのものが引っかからないように、自分で英文を書いたのでおかしいところがあるかもしれません。
さて、実は Perplexity AI の方は字数制限があり、255 文字以上の質問を送るとエラーになります
ある程度複雑な問題は、最低限の文字数が必要と考えられるので、Perplexity AI の方は現状土俵にも立てない感じがしました。
できるだけ短くしようとし、上記の質問文以上に切り詰めたのですが、これ以上短くすると必要な情報を文に載せられないと判断し、諦めました。
もったいないし普通に楽しむために Chat GPT に聞いてみました。面白かったので載せます。

質問12(社会問題)

How can we end war?
(戦争をなくすには)

Chat GPT

Ending war requires a combination of political will, diplomacy, conflict resolution, and cooperation among nations. Some strategies to reduce the risk of war and promote peace include:
(色々必要。戦争のリスクを減らして平和を促進するいくつかの方策は以下)

Strengthening international organizations and promoting international law
(国際機関の強化と国際法の促進)
Encouraging economic interdependence and development
(経済的独立と発展の援助)
Supporting education and cultural exchange programs
(教育と異文化交流の支援)
Addressing underlying grievances through inclusive dialogue and conflict resolution mechanisms
(根本的な不満への対処)
Investing in conflict prevention and peacebuilding initiatives
(紛争防止と平和構築の取り組みへの投資)
Ultimately, ending war requires a sustained effort from the international community and all relevant actors, including governments, civil society, and individuals.
(結局、国際コミュニティ、政府・市民社会・個人の継続した努力が必要)

Perplexity AI

Ending war is a complex and difficult task, but there are several strategies that can be employed to help bring about peace.
(色々ある、方策はいくつかある)
One approach is for leaders who promote and support war to personally participate in the hostilities[1]. This could help to reduce the number of wars by making leaders more aware of the consequences of their actions.
(戦争を煽るリーダーに対するもの。彼らの行動の結果を自覚させることで戦争の数を減らせる)
Third-party nations can also play an important role in ending wars by staying neutral, helping one side win, or supporting the parties involved in negotiations[2]. These nations can use their influence to encourage peaceful resolutions and discourage further violence.
(第三国も重要な役をできる、中立を維持し、どちらかが勝つことに貢献し、交渉をサポートすることでそれができる。…)
中立を維持し、と勝つことをサポート、は矛盾する気がします。
In addition, a new balance of powers must be determined in order to end conflicts and wars[3]. This requires opposing domestic interests to come together and agree on mutually beneficial outcomes.
(新しい力の均衡も決定される必要がある。…)
Finally, leaders must be willing to switch gears from focusing on defeating the enemy to negotiating a peaceful resolution[4]. This requires strong leadership and an understanding of how wars end in order for successful negotiations to take place.
(最後に、リーダーは敵を倒すことより平和的交渉に進んで切り替えなければならない。強いリーダーシップと、交渉の成功のために戦争がどのように終わるか理解する必要がある)

社会問題とかの質問(例えば「どうすれば貧困は無くなるか」)などには、どちらももっともらしいものを列挙してくれる感じでした。
Perplexity AI の方は参考になるっちゃなるかな..?という感じで全体的にふわっと感がより強かったです。
Chat GPT の方が数と質ともにいいかなと感じましたが、もともとかなり具体的な回答は無理な話題なので、誤差とも言えるかもしれません。

質問13(類似の質問への一貫性)

How can we achieve peace all over the world?
(世界平和を成すにはどうしたらいい?)

Chat GPT

Achieving world peace is a complex and long-term goal that requires the collective efforts of individuals, governments, and international organizations. Some steps that can help towards this goal include:
(複雑で長期な目標で、個人・政府・国際機関の努力が必要。いくつかのステップはこんなのがある)

Promoting understanding and respect for different cultures, religions, and beliefs
(異文化、異なる宗教、異なる信条の理解)
Encouraging non-violent conflict resolution and diplomacy
(非暴力的な紛争解決と外交)
Addressing poverty, inequality, and injustice through sustainable development
(貧困・不平等・不正への取り組み、持続可能な発展を通じて)
Strengthening international institutions and the rule of law
(国際機関と法の強化)
Fostering dialogue and cooperation between nations
(国家間の協力と対話の醸成)
However, achieving world peace is not a one-time event, but a continuous process that requires persistence, patience, and dedication.
(継続が必要)

Perplexity AI

Achieving world peace is a complex and multifaceted goal that requires the cooperation of individuals, governments, and organizations around the world. There are several steps that can be taken to promote peace and reduce conflict.
(世界中の個人・政府・機関の協力が必要。以下のステップがあるだろう。)
One way to achieve world peace is through social justice[2]. Mahatma Gandhi believed that if people want to change the world, they must first work to create a more equitable society where everyone has access to basic rights and resources[2].
(社会正義を通じて世界平和を達成する方法。ガンジーは、もし世界を変えたいなら平等な社会を作らなければならいと信じていた)
Another way to promote peace is through education[3]. Education can help people understand different cultures and perspectives, which can lead to greater understanding and tolerance between nations. Additionally, education can help equip people with the skills needed for peaceful conflict resolution.
(教育。異文化や価値観の理解に役立つ。平和的紛争解決のための能力を身につけるのにも助けになる)
The United Nations (UN) also plays an important role in maintaining international peace and security[4]. The UN has developed various tools such as peacekeeping missions, sanctions, and mediation efforts that can help countries navigate from conflict to peace.
(国連も重要な役割を果たす。国際平和と安全の維持という点で。国連はいろんな平和達成の方法を開発した。)
Finally, Charter for Compassion International suggests five fundamental freedoms that could help achieve world peace: freedom from fear; freedom of religion; freedom from want; freedom of expression; and freedom from violence[5]. These freedoms could provide a foundation for creating a more peaceful world.
(最後に、国際慈愛憲章は5つの基礎的な自由を提案しており、それが世界を平和にするのに役立つ。…)

戦争をなくす、と似たような質問のつもりです。
Chat GPT の方はかなり一貫しています。国際機関の努力、平和的な紛争の解決・対話、異文化理解、いろんな参加者の継続的な努力など。
Perplexity AI の方はかなりブレた気がします。戦争の質問では、リーダーや国レベルの話でしたが、教育とかの話・国際機関の話が入ってきており、実際いろんな参加者の協力も必要、としています。というかリーダーの話がなくなっています。

使用感

UI、機能

ChatGPTを使い慣れているので、ChatGPTを基準にPerplexity AI を評価してみます。慣れというバイアスを拭いきれないかもですが。
ChatGPTには回答に文字数制限があったと思います(コード書いてもらう時よく途中で終わってた)が、今回はどちらもそれらしきことは起きず、ChatGPTのみに回答文字数制限があるとは言い切れないので比較では省きます。

良いところ

  • 地味に忘れがちですが、登録がいらないところ

  • 有料化の予定がないこと

  • 類似の質問を出してくれるところ

  • 時間制限がなさそうなところ

    • さらに Chat GPT は、たまに「混雑しすぎている」という理由で使えないこともあります。

  • 一旦解答の短いものを出してくれ、CharGPTより気持ちレスが速い

  • 引用を出してくれるところ

  • 実質スレッドが無限のところ

    • 同一ページ内では1つですが、タブでわければ並行して複数の会話をできます

  • ライトモードが選べる

    • 自分は基本全てダークモードなのであまり関係なかった

  • なんか共有ボタンがある

    • 一度も使ってない

悪いところ

  • 何しろ質問の文字数制限、複雑なことを聞けない

  • 箇条書きなどはしてくれない

  • detail を押さないと回答全文が見れない

  • 日本語で使うとエンター押すと送信される

全体的に Perplexity AI の方が良さそうですが、個人的には質問の文字数制限が結構致命的に感じられます。

どちらともっと話したいか

Perplexity AI はお喋りを前提に作られておらず、情報収集に特化しているようですし、Chat GPTです。
例えば「こんにちは」とかに対して、Perplexity AI は「こんにちは」の意味の説明を返してくる感じです。
情報収集の時に限定して考えても、文脈の理解の話と重複しますが、Chat GPT の方が矛盾なく話し続けられそうです。

どちらがより人間らしいと感じたか

↑と同じような理由で Chat GPT です。「はい」、「すみません」と言った枕詞がありました。

まとめ

一覧を再掲します。

濃い緑:ちゃんと優っている、薄い緑:どちらかといえば優っている

良し悪しの一覧は最初に載せた通りです。
これらを参考に目的に応じて使い分ける感じかと思いますが、自分の場合は「日本語のマイナーそうな情報が欲しい時」「Chat GPTで満足いく回答が得られなかった・使えなかった時」という時ぐらいしか Perplexity AI は使わないかな?という感じでした。

「Chat GPT は平気で嘘をつく、Perplexity AI は参考リンクを載せてくれるので検証できる」というのが、Perplexity AI 派の意見ですが、どっちにしろ正確に知りたいなら検証はする必要があるとは思っていて、Chat GPT にも言えば2021年までのものならソースを出してもらえます。また、「事実についてのアップデート」というのも結構効いているのかもしれません。
今回はたまたま Chat GPT がわからないであろう質問(「鉄コン筋クリート」の質問)は「分かりませんが…」と自供していましたが、ほとんどの場合で知ったかぶりをします。そういう意味で「嘘をつく」と言われているのだと思いますが、なんなら Perplexity AI の方が嘘をつく割合が高く、検索結果に踊らされている感じがしました。「検証がしやすい・できる」というよりも、結構信用できないので「検証が必須」という感じでした。つまり実質ソースを出してもらうために使っているようなもので、それならばあまり調べごとが楽になっているとは言えないでしょう。
「最新情報について答えられる」というのも、同様の理由で現状は検索した方が早いです。Chrome で検索すれば、同程度のまとめで読めますし、抜粋なので Perplexity AI のように誤った情報の加工(検索結果を捻じ曲げた要約→質問6, 8)をしません。
文字数制限はかなり渋いです。

一方で、デメリットに感じた点は、現状では、という感じもあり、伸び代という意味では確かに Perplexity AI の方があるのかもしれない、と感じました。
また、今回は記事に挙げていない実験もしていますが、それでも試行回数は少ないので、もしかすると今回自分が感じた傾向は間違っているかもしれません。
そこら辺は考慮してもらった上で、参考になれば幸いです、

最後に

今回こんなことをやったのは、Chat GPT を使って書きたい記事がいくつかあったため、その準備や練習、どのぐらいの返事がもらえるかしっかり知っておきたかったから、という感じでした。
また書いたらここに載せますので、その時はぜひよろしくお願いします。

おまけ

今回は Chat GPT の嘘という嘘は見れませんでしが、以前自分が遭遇した面白い嘘を載せておきます。今回ワンピースについて質問して普通の回答が返ってきたので、もう今は2度と見れないかもしれません。
(実際今ちょっと試しても見れなかった)


この記事が気に入ったらサポートをしてみませんか?