Name: 你的 AI 夠聽話嗎？快速提升 AI 技能的 3 大關鍵（How to boost your AI skills ? | BBC Ideas）
Uploaded: 2026-03-05T17:37:53.000Z
Duration: 4 min 44 s

情態副詞

So you've probably already started using AI chatbots,

你可能已經開始使用 AI 聊天機器人了，

maybe to create a meal plan or write a difficult email.

也許是用來制定飲食計畫或寫一封難搞的郵件。

If you haven't, check out our handy intro guides at bbc.co.uk/ideas or OU Connect.

如果你還沒試過，可以到 BBC Ideas 或 OU Connect 查看我們的入門指南。

But you're not sure you're getting the best out of it,

但你不確定自己是否發揮了它的最大效用，

and it hasn't exactly changed your life yet.

而且它似乎還沒真正改變你的生活。

那麼，該如何提升你的使用技巧呢？

This short guide will cover how to write better prompts,

這本簡短指南將涵蓋如何寫出更好的提示詞（Prompts）、

how to avoid something called "AI sycophancy,"

如何避免所謂的「AI 諂媚現象（AI Sycophancy）」，

and how to pick the right chatbot for your needs.

以及如何根據你的需求選擇合適的聊天機器人。

First, you need to appreciate that AI is a machine—a mathematical model—

首先，你得明白 AI 是一台機器、一個數學模型，

When you're giving it instructions or prompts,

you're guiding it to give you answers, not asking a magician to conjure up a miracle.

你是在引導它給出答案，而不是要求魔術師變出奇蹟。

It can help to think of your chatbot as a small spaceship

你可以把聊天機器人想像成一艘小型太空船，

which follows your instructions to find the answers from the vast galaxy of interconnected information that's out there.

它根據你的指令，在龐大且相互關聯的信息銀河中尋找答案。

Say you're looking for advice as to what type of dog your family should get.

假設你正在尋求建議，想知道家裡適合養什麼品種的狗。

If you type the question into Google, you'll get results from dog food companies, charities, even banks;

如果你在 Google 輸入這個問題，會得到來自狗糧公司、慈善機構甚至銀行的結果，

and some will have paid Google to showcase their results when that question is typed in.

其中有些是付錢給 Google，讓他們在有人搜尋時展示自己的結果。

If you put the same question into an AI chatbot,

如果你把同樣的問題丟給 AI 聊天機器人，

you'll get a long list of different dogs, unlikely to be specific to your needs.

你會得到一長串不同的狗，但不太可能完全符合你的需求。

In fact, it doesn't even know what a dog is.

事實上，它甚至不知道什麼是「狗」。

In the AI information galaxy, "dog" is next to "puppy," "bone," and "walk."

在 AI 的信息銀河中，「狗」與「幼犬」、「骨頭」和「散步」相連。

This demonstrates how the AI understands and connects concepts.

這展示了 AI 如何理解並連結概念。

So, if you make the connections for it—in other words, give it the right context—

所以，如果你幫它建立連結——換句話說，給它正確的語境（Context）——

it's much more likely to produce a useful response.

它就更有可能產出有用的回應。

"Help me decide which dog would make a good pet for my busy family with three teenage children, two cats, and a small garden,

「請幫我決定哪種狗適合我這個忙碌的家庭：有三個青少年、兩隻貓和一個小花園，

請參考位於英國的可信來源。」

Essentially, you're helping it narrow down the vast array of information it can access

本質上，你是在幫它縮小海量信息的範圍，

and navigate to a much more specific bit of the information galaxy.

引導它航向信息銀河中更具體的區域。

Another tip to get better results from your interactions with AI is to think about your tone when writing your prompts.

另一個獲得更好結果的技巧是，在寫提示詞時注意你的語氣。

If you're polite and friendly, you're likely to get polite and friendly responses back,

如果你禮貌且友好，通常會得到禮貌且友好的回應，

because you've steered the spaceship to the "friendly word" section of the galaxy.

因為你把太空船駛向了銀河中的「友好詞彙區」。

If you type an angry prompt, you steer the spaceship into the "angry sector" of the galaxy.

如果你輸入憤怒的提示詞，你就把太空船駛進了銀河中的「憤怒區」。

The AI isn't becoming angry; it doesn't really have a personality.

AI 並不會真的變生氣，它並沒有性格。

You just steered it to where the angry words live.

你只是把它引導到了憤怒詞彙聚集的地方。

Sycophancy can sometimes be a problem when using chatbots.

使用聊天機器人時，「諂媚」有時會成為一個問題。

That's the AI telling you what it thinks you want to hear.

那是 AI 在說它認為你想聽的話。

This could be awkward at work if you're trying to improve a business process or come up with a new idea,

如果你正在嘗試優化業務流程或提出新點子，這在工作中可能會很尷尬，

and it's constantly telling you that your ideas are brilliant.

因為它會不斷告訴你你的主意棒極了。

One tip to counteract that is to specifically ask the chatbot for different perspectives.

對抗這種情況的一個技巧是，明確要求機器人提供不同的觀點。

Prompt it with: "Play Devil's Advocate," or "Are there counter-arguments to moving to a four-day week?"

提示它「請扮演反方角色（Devil's Advocate）」或「實施週休四日是否有反面論點？」

Asking the chatbot to explain its reasoning and include sources can also be useful,

要求聊天機器人解釋其推理過程並列出來源也非常有用，

so you can see the journey it's been on and how it's reached its conclusions.

這樣你就能看到它的思考路徑以及它是如何得出結論的。

還有「幻覺（Hallucinations）」問題。

AI 會給出聽起來很有說服力但卻錯誤的答案，

because they've wrongly interpreted something in a bit of the information galaxy they're searching in.

因為它錯誤地解讀了正在搜尋的信息銀河中的某些內容。

They're getting rarer, but they're absolutely still out there.

這種情況雖然越來越少見，但絕對仍然存在。

So, it's important to fact-check AI responses by comparing them to reliable sources you trust.

因此，透過與你信任的可靠來源進行對比，來查證 AI 的回應是非常重要的。

And think carefully about which chatbot you're using.

並且要仔細考慮你使用的是哪款聊天機器人。

Different models have been trained on different material and may have different biases.

不同的模型使用不同的材料進行訓練，可能會有不同的偏見。

So a model like Grok, the AI developed by X,

is likely to give you a more sarcastic answer than one developed for a corporate client by a firm like Microsoft or Google.

給出的答案可能比微軟或 Google 為企業客戶開發的模型更具諷刺意味。

Be aware of the impact of so-called "guardrails" on results.

要注意所謂的「護欄（Guardrails）」對結果的影響。

For example, some AIs are designed with safety in mind, which could be helpful,

例如，有些 AI 的設計考量了安全性，這很有幫助，

like warning you not to self-harm or download illegal content.

比如警告你不要自殘或下載非法內容。

But guardrails could be downright annoying if, for example,

但「護欄」有時也會讓人惱火，例如，

you're trying to research the 1989 Tiananmen Square protests using a Chinese chatbot which has censorship built in.

如果你試圖使用內建審查機制的中國聊天機器人來研究 1989 年的天安門事件。

In short, if you're going to get the best out of AI, use it carefully.

簡而言之，如果你想發揮 AI 的最大效用，請謹慎使用。

Think about what you're asking it to do and how.

思考你要求它做什麼以及如何要求。

It can be great as a tool for brainstorming and support in making decisions,

它作為激盪靈感和輔助決策的工具非常出色，

but it's not a replacement for critical thinking.

If you understand its mechanics, limitations, and biases,

如果你了解它的機制、局限性和偏見，

you'll treat it as a useful tool, not a magician.

你就會把它當作一個有用的工具，而不是魔術師。