善用 ChatGPT：避免掉入自己的陷阱

 20240825

我是台灣高鐵的常客，在候車時經常會看牆上的安全宣導影片。因為高鐵的中譯英是我長久以來英語教學的教材之一。

有天，發現一則提醒：「請注意您的隨身行李以防遭竊」，其英譯為「Please be aware of your belongings to avoid stealing」。乍看，直覺這個翻譯有點不妥，正確的英文表達應該是「avoid theft」。「Avoid stealing」意指防止自己偷竊，顯然與原意大相逕庭。

出於好奇和對這個翻譯的錯誤，我決定測試一下我的兒子。我的兒子，對英語和 ChatGPT 同樣很感興趣。

我問他這句英文「Please be aware of your belongings to avoid stealing」是否誤譯。令人驚訝的是，他說 ChatGPT 也是將此句英文翻譯成：「請注意您的隨身行李以防遭竊」。也因此，他質疑我為什麼懷疑「Avoid stealing」是誤譯呢。

這個事件，證實了一般人們的信念 — 認為像 ChatGPT 這樣功能強大的 AI 生成的翻譯應該是沒問題的。

將「以防遭竊」翻譯成「to avoid stealing」的問題，可能源於我們目前的英語教學在區分動詞、現在分詞、動名詞和名詞方面，仍存在一定的挑戰和不足。

而且，這一事件引起我更深入思考，我們應該如何與 ChatGPT 這樣強大的工具互動。我所關心的不僅僅是翻譯本身這個問題，而是人們可能過度依賴 AI 而沒有充分了解其限制的潛在風險。

ChatGPT 確實是一個非凡的工具，能夠理解並回應來自世界各地使用者的廣泛輸入。然而，這種能力伴隨著一個警告：如果不質疑或完全理解 AI 的輸出原理，我們可能會無意中陷入自己設下的陷阱。

此句高鐵中文英譯的問題，提醒我們必須謹慎運用 ChatGPT。正確的詢問方式應該是：「請問 ‘Please be aware of your belongings to avoid stealing’ 這個句子是否正確？如果不正確，請修正，再翻譯成中文。」這樣提問方能確保英文原句的準確性，從而獲得正確的翻譯。

正如所謂「水能載舟，亦能覆舟」，功能強大的 ChatGPT 能幫助我們，也可能使我們誤入歧途。

何況，ChatGPT 並非設計用來接受像「Please be aware of your belongings to avoid stealing」這樣非正式或不正確表達的訓練，但它仍能盡力解讀使用者提供的提示或問題，即使句法有誤，也能產生正確的回應。

這種能力是 ChatGPT 模型設計中的一個關鍵特性。然而，這也突顯了我們與 ChatGPT 之間溝通的重要性，以及不加批判地接受 AI 輸出的風險。

高鐵翻譯的故事以及我兒子與 ChatGPT 互動的過程和結果提醒我們，為有效駕馭及運用 ChatGPT，必須先徹底了解其生成原理及正確的使用方法，以發揮其最大的功能。

期望在 ChatGPT 陪伴我們的旅途中避免掉入自己無意中的陷阱。

Making the Most of ChatGPT: Avoiding the Traps You Set for Yourself

I am a frequent user of Taiwan’s High-Speed Rail, and I often watch the safety videos displayed on the walls while waiting for my train because the HSR’s Chinese-to-English translations have long been one of my teaching materials for English. One day, I noticed a reminder: “請注意您的隨身行李以防遭竊,” which was translated into English as “Please be aware of your belongings to avoid stealing.” At first glance, I instinctively felt that this translation was somewhat incorrect; the correct English expression should be “avoid theft.” “Avoid stealing” implies preventing oneself from committing theft, which is clearly far from the intended meaning.

Out of curiosity and concern about this translation error, I asked my son, who is equally interested in English and ChatGPT, whether the sentence “Please be aware of your belongings to avoid stealing” was mistranslated. To my surprise, he replied that ChatGPT also translated this sentence as “請注意您的隨身行李以防遭竊.” Because of this, he questioned why I would doubt that “Avoid stealing” was a mistranslation. This incident confirmed a common belief—that a powerful AI like ChatGPT must produce correct translations.

The issue of translating “以防遭竊” as “to avoid stealing” might stem from the challenges and shortcomings in our current English education system when it comes to distinguishing between verbs, present participles, gerunds, and nouns. Moreover, this incident prompted me to think more deeply about how we should interact with a powerful tool like ChatGPT. My concern is not just about the translation itself but also about the potential risks of people over-relying on AI without fully understanding its limitations. ChatGPT is indeed an extraordinary tool, capable of understanding and responding to a wide range of inputs from users all over the world. However, this ability comes with a warning: if we do not question or fully understand the principles behind AI’s outputs, we might inadvertently fall into traps of our own making.

This issue with the High-Speed Rail’s Chinese-to-English translation reminds us that we must use ChatGPT cautiously. The correct way to ask would be: “Is the sentence ‘Please be aware of your belongings to avoid stealing’ correct? If not, please correct it and then translate it into Chinese.” Asking in this way ensures the accuracy of the original English sentence, leading to a correct translation.

As the saying goes, “Water can carry a boat, but it can also overturn it.” Similarly, the powerful ChatGPT can assist us, but it can also lead us astray. Moreover, ChatGPT is not designed to be trained on informal or incorrect expressions like “Please be aware of your belongings to avoid stealing,” yet it still strives to interpret the prompts or questions provided by users, and even when there are syntactical errors, it can still produce correct responses. This capability is a key feature of how ChatGPT is designed. However, this also highlights the importance of how we communicate with ChatGPT and the risks of uncritically accepting AI outputs.

The story of the High-Speed Rail translation and the interaction between my son and ChatGPT remind us that to effectively navigate and utilize ChatGPT, we must first thoroughly understand its underlying mechanisms and the correct way to use it to maximize its potential. We hope to avoid falling into unintentional traps of our own making as ChatGPT accompanies us on our journey.