Name: 12分鐘教你搞定！你唯一需要的AI工具們！ (The Only AI Tools You Need (12-Minute Guide))
Uploaded: 2026-01-26T04:36:14.000Z
Duration: 11 min 56 s

I use around 10 AI tools for 90% of my work, and

我大概有九成的時間都用十幾個 AI 工具，

But figuring out which tool works best for what task usually takes months of trial and error.

但要搞清楚哪個工具適合做什麼，通常都要花好幾個月試了又試。

So I'll share the one thing each tool does better than alternatives, so you walk away with a clear mental model for when to use what.

所以我會分享每個工具最厲害的地方，讓你一聽就懂，知道什麼時候該用哪個。

I've grouped these tools into four categories across a two part series.

我把這些工具抽成四大類，做成兩集的系列。

This video covers everyday and specialist AI.

這支影片會講日常和專業用的 AI。

While part two covers the remaining two categories, let's get started kicking things off with everyday AI.

而第二集會講剩下的兩類，那我們就先來聊聊日常用的 AI 吧。

These are your general purpose chatbots, ChatGPT, Gemini, and Claude.

這些就是你常用的哈拉機器人，像是 ChatGPT、Gemini 和 Claude。

And while they seem interchangeable, their motes the specific things they do best have actually become quite distinct, starting with the OG ChatGPT.

而雖然它們看起來可以互相替換，但它們最擅長處理的具體事情其實已經變得相當不同，從最經典的 ChatGPT 開始。

While Gemini and Claude are arguably just as capable in raw power, ChatGPT still holds the crown in one area.

雖然 Gemini 和 Claude 在原始能力上可以說同樣強大，但 ChatGPT 在某個領域仍然是王者。

It's in plain English, ChatGPT drops fewer balls when you hand it a complex checklist.

簡單來說，當你給 ChatGPT 一個複雜的檢查清單時，它比較不容易出錯。

give them a lengthy set of instructions and they'll sometimes skip a step or decide they know better.

當你給它們一長串訓示時，它們有時會跳過步驟，或者自以為知道得比較多。

If you want proof of this, just ask each model to optimize a rough prompt for itself.

如果你想證明這一點，就讓每個模型為自己優化一個粗略的 prompt。

ChatGPT will generate a noticeably longer and more detailed prompt, because it knows it can handle the complexity.

ChatGPT 會生成一個明顯更長、更詳細的 prompt，因為它知道自己可以處理這種複雜性。

And if you run that optimized ChatGPT prompt through both ChatGPT and Gemini, for example, you'll notice two things.

而且，如果你將優化後的 ChatGPT prompt 輸入到 ChatGPT 和 Gemini 中，你會注意到兩件事。

First, ChatGPT thinks longer because it's actually checking every requirement, and it follows each instruction to the letter.

首先，ChatGPT 會思考更久，因為它實際上在檢查每一個要求，並且會逐字逐句地遵循每個訓示。

Gemini, on the other hand, often takes shortcuts.

Pro tip I share the exact prompt optimizer in the Essential Power Prompts template linked below, but

小提示：我會在下方連結的 Essential Power Prompts 範本中分享確切的 prompt optimizer，但

you can test this yourself with something as simple as optimize this prompt for ChatGPT.

你也可以自己測試，用「優化這個 prompt 給 ChatGPT」這樣簡單的指令。

Diving into a real world example, I gave both ChatGPT and Gemini the same complex prompt, a hiring rubric with a dozen requirements.

進入實際的例子，我給了 ChatGPT 和 Gemini 同一個複雜的 prompt，一個有十二項要求的招聘標準。

Gemini's output looked right at first glance, but

Gemini 的輸出乍看之下沒問題，但

when I checked it against my original list, it had quietly dropped a few rules.

當我對照原始清單檢查時，它卻悄悄地省略了幾項規則。

ChatGPT doesn't decide which instructions matter, it just follows them.

ChatGPT 不會自己判斷哪些指令比較重要，它就是照著做。

這裡還有第二個比較簡單的例子。

Sometimes when you explicitly tell Gemini to search the web, it just doesn't.

有時候，就算你明確告訴 Gemini 要做網路搜尋，它就是沒做。

Which is wild, since Gemini and Google Search are both Google products, right?

這很扯，因為 Gemini 和 Google Search 都是 Google 的產品，對吧？

Whereas with ChatGPT, when you enable web search, it performs the web search every single time.

相較之下，用 ChatGPT 的話，只要你啟用網路搜尋，它每次都會執行網路搜尋。

I know this is a small example, but it's downstream from ChatGPT's core superpower.

我知道這只是個小例子，但這其實源自於 ChatGPT 的核心超能力。

Obedience means you can trust the behavior you ask for.

聽話代表你可以信任你要求的行為。

So as a rule of thumb, if the task has a lot of moving parts and getting one wrong breaks the whole thing.

所以，有個經驗法則，如果任務有很多細節，而且其中一個弄錯就會讓整個事情失敗。

Next up, Gemini where ChatGPT wins on obedience, Gemini wins on multimodality.

接下來是 Gemini，雖然 ChatGPT 在聽話方面贏了，但 Gemini 在 multimodality 方面贏了。

In plain English, Gemini is able to process a massive amount of mixed media video, audio, images and text natively.

簡單來說，Gemini 原生就能處理大量的混合媒體，像是影片、音訊、圖片和文字。

Taking a look at this table, we see that only Gemini can handle all four types of media natively.

看看這個表格，我們可以看到只有 Gemini 能原生處理這四種媒體。

It's able to quote unquote, listen to audio and quote unquote watch videos while ChatGPT and Claude use roundabout ways to access that information.

Gemini 可以「聽」音訊和「看」影片，而 ChatGPT 和 Claude 則是透過迂迴的方式來存取這些資訊。

What's more, Gemini's massive 1 million token context window means it can handle large video recordings, hour long audio recordings, full slide decks all together that would literally choke other models.

更厲害的是，Gemini 擁有龐大的 100 萬 token 上下文視窗，這代表它可以一次處理大型的影片錄影、長達一小時的音訊錄影、完整的投影片組，這些其他模型可能根本跑不動。

If you watch my latest Gemini video, you remember the use case where I screen recorded a messy walkthrough of myself completing a task, uploading that video onto Gemini and asking Gemini to turn it into a ready to use SOP with perfect formatting, which is an example of Gemini ingesting video and turning it into text.

如果你看過我最新的 Gemini 影片，你應該還記得那個案例，我錄下了自己完成任務的畫面，然後上傳到 Gemini，請 Gemini 把影片轉成格式完美的 SOP，這就是 Gemini 讀取影片並轉成文字的例子。

現在我們來把這件事做得更深入一點。

Imagine you just finished a weekly meeting.

想像一下，你剛開完一個每週例會。

You have a video recording of the call, a 20 slide deck and a and a photo of a messy whiteboard session.

你有會議的錄影、一份 20 頁的投影片，還有一張畫滿雜亂手稿的白板照片。

You can upload all three and ask Gemini to summarize what was discussed, pull out the key decisions and draft the follow up email.

你可以一次上傳這三樣東西，然後請 Gemini 總結討論內容、整理出關鍵決策，並草擬後續的電子郵件。

Gemini is the only tool that can synthesize all three in one go.

Gemini 是唯一一個能一次整合這三種資訊的工具。

All that said, I have to point out that Gemini's raw reasoning capabilities sometimes feels slightly behind ChatGPT, but

不過話說回來，還是得說 Gemini 的純粹推理能力有時候感覺比 ChatGPT 稍微弱一點，但

when the task involves video, audio or massive files, the trade off is obviously worth it.

如果任務涉及到影片、音訊或超大型檔案，這個取捨顯然是值得的。

Speaking of matching the right tool to the task, today's sponsor HubSpot put together a free guide called the AI Productivity Stack that covers 50 tools organized by use case.

說到要為任務找到對的工具，今天的贊助商 HubSpot 整理了一份免費指南，叫做「AI 生產力工具堆疊」，裡麵包含了 50 種依據使用情境分類的工具。

While this video focuses on my personal favorites, your workflow probably needs something different.

雖然這支影片主要講我個人最喜歡的工具，但你的工作流程可能需要不一樣的東西。

Maybe you're in marketing and need SEO specific tools, or you manage a team and want to build automated workflows with reliable AI.

也許你是做行銷的，需要專門的 SEO 工具；或者你管理一個團隊，想用可靠的 AI 建立自動化工作流程。

This guide breaks down tools across business functions like research, design and marketing, and

這份指南將工具依據研究、設計和行銷等商業功能做分類，

for each tool it shows you the best use case, key features, pricing, and

而且針對每一種工具，它會告訴你最佳使用情境、主要功能、定價，以及

一個步驟接著一個步驟的工作流程。

What I found most useful is the decision logic at the end of each.

我發現最有用的部分是每一項工具最後的決策邏輯。

So for example, the research category tells you exactly when to use Perplexity versus Claude versus Humada based on what you're actually trying to do, it's a great way to quickly understand what each tool does well.

所以舉例來說，研究類別會告訴你，根據你實際想做的事，到底該用 Perplexity、Claude 還是 Humada，這是一個能快速瞭解每個工具擅長之處的好方法。

I'll leave a link to this free guide down below.

我會把這份免費指南的連結放在下方。

Thank you HubSpot for sponsoring this video.

Rounding out the everyday AI category Claude Claude Superpower is producing higher quality first drafts than the other models.

在日常 AI 類別中，Claude 的超能力是能產出比其他模型更高品質的初稿。

In plain English, that means Claude's first attempt is usually closer to done.

簡單來說，這意味著 Claude 的第一次嘗試通常已經接近完成了。

首先是程式碼，這裡有個有趣的點。

The latest version of Gemini beat the older version of Claude in every single benchmark score, except for the coding one, which is crazy.

最新版本的 Gemini 在所有基準測試分數上都擊敗了舊版本的 Claude，唯獨在程式碼方面沒有，這真的很瘋狂。

So obviously Anthropic has figured out something related to coding, the others haven't.

所以顯然 Anthropic 在程式碼方面已經掌握了某些訣竅，而其他公司還沒有。

And in practice, developers universally agree that Claude writes functional code on the first try more consistently than alternatives.

而且實際上，開發者普遍認為 Claude 比其他替代方案更能穩定地在第一次就寫出可用的程式碼。

I needed to bulk export conversations from a customer service platform, but

我需要從一個客戶服務平臺大量匯出對話紀錄，但是

the support team said only developers could do it.

支援團隊說只有開發者才能做到。

I described the problem and Claude not only gave me step by step instructions, but

我描述了問題，Claude 不僅給了我步驟說明，還

also wrote a script in Go that worked on the first try.

寫了一個 Go 腳本，第一次就成功了。

I don't even know what Go is, nor can I write code.

我根本不知道 Go 是什麼，也寫不出程式碼。

Another example I asked all three models to turn a static image into an interactive chart, and

另一個例子是，我請這三個模型將一張靜態圖片轉換成互動式圖表，

Claude performed the best on the first try.

Claude 在第一次嘗試時表現最好。

So basically, anything that requires generating working code tends to favor Claude Pro Tip when it comes to diagrams, you can ask Claude to generate mermaid code, which you can then paste directly into tools like Excalidraw to get clean visuals in minutes.

所以基本上，任何需要生成可用程式碼的東西，通常都偏好 Claude。小提示，關於圖表，你可以請 Claude 生成 mermaid code，然後直接貼到 Excalidraw 這類的工具裡，幾分鐘內就能得到乾淨的視覺圖。

Area 2 Polishing copy beyond Code Claude produces written drafts that sound human and need fewer revisions.

第二個部分：潤飾文案，超越程式碼。Claude 生成的文案草稿聽起來很像真人寫的，需要較少的修改。

When you need to tighten an argument or match a specific voice, Claude just gets it.

當你需要讓論點更精煉，或是符合特定語氣時，Claude 都能完美掌握。

Put simply, it's exceptionally good at style matching.

簡單來說，它在語氣風格的匹配上表現非常出色。

Once you share examples of your existing work, it replicates your tone almost perfectly.

只要你分享現有作品的範例，它幾乎能完美複製你的語氣。

When I was in corporate, I'd shared previous documents so Claude could replicate that voice across presentations and performance reviews.

我以前在公司上班時，會分享過去的檔案，讓 Claude 可以在簡報和績效評估中複製那樣的語氣。

And now as a creator, I feed it my existing YouTube scripts to help refine new drafts.

現在身為創作者，我會把現有的 YouTube 腳本餵給它，來幫助我潤飾新的草稿。

At this point, you might be wondering how I use all three everyday AI tools together.

這時候，你可能會好奇我平常是怎麼一起使用這三個 AI 工具的。

In a nutshell, ChatGPT or Gemini usually handles the beginning of my work ideation research, drafting the outline of a presentation Claude that handles the last mile, turning that rough output into something I'm ready to present or publish.

總之，ChatGPT 或 Gemini 通常會負責我工作發想研究的開頭，還有簡報大綱的草擬，而 Claude 則負責最後一哩路，將那些粗糙的產出變成我準備好要呈現或發表的內容。

Quick note on Grok A lot of people ask why I don't use it.

關於 Grok 的小提示：很多人問我為什麼不使用它。

Grok's superpower is its direct access to the Twitter x Firehose right?

Grok 的超能力是它能直接存取 Twitter x Firehose，對吧？

So it's the best option for people who need to analyze breaking news events in real time.

所以對於需要即時分析突發新聞事件的人來說，這是最好的選擇。

And as a rule of thumb, we should never use tools just for the sake of using tools.

一般來說，我們絕對不該為了用而用工具。

We should only add them to our toolkit when they solve an actual problem.

我們應該只在工具能解決實際問題時，才將它們加入我們的工具箱。

We have Here's a quick recap of the three models and when to use them.

這裡快速總結一下這三個模型以及何時該使用它們。

And if you're wondering whether you need all three, the short answer is no.

如果你在想需不需要這三個，簡單來說就是不用。

Most people should stick with a paid version of ChatGPT and get really good at it.

大多數人應該都只需要付費版的 ChatGPT，並且把它用得很熟練就好。

But if you can afford multiple subscriptions and your workflow can take advantage of their individual superpowers, mix and match as needed.

但如果你負擔得起多個訂閱，而且你的工作流程能發揮它們各自的超能力，那就視情況混合搭配使用。

Fun fact According to this study on open router data models from different labs like ChatGPT and Gemini and expand the pie of AI use cases precisely because they excel at different things onto the second category, specialist AI.

有趣的事實 根據這份關於來自 ChatGPT 和 Gemini 等不同實驗室的開放路由數據模型的研究，透過它們各自擅長的不同事物，擴展了 AI 用例的範圍，精準地進入了第二類，也就是專家級 AI。

Before diving in, let's clear up a very common misconception.

在深入探討之前，我們先釐清一個非常常見的誤解。

字幕列表影片播放

12分鐘教你搞定！你唯一需要的AI工具們！ (The Only AI Tools You Need (12-Minute Guide))

specific

entire

massive

context