# 策展 · X (Twitter) 🔥🔥🔥🔥🔥

> 📖 本站完整內容索引（documentation index）：[llms.txt](/llms.txt)

> 作者：Sam Altman (@sama) · 平台：X (Twitter) · 日期：2026-07-09

> 原始來源：https://x.com/sama/status/2074909079450050629

## 中文摘要

OpenAI 在 ChatGPT 推出 GPT-Live 支援即時語音對話。

<video src="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1783562614324-1v3k03xm.mp4" poster="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/535f765f9cf733cc.jpg" controls playsinline preload="metadata" style="max-width:100%;height:auto;display:block;margin:1rem 0"></video>
> 影片展示了 ChatGPT 應用程式中 GPT-Live 的語音互動介面。

**核心功能與架構**
Sam Altman 指出，GPT-Live 的推出讓與 AI 的對話變得極具真實感與「魔法」，甚至讓他改變了以往偏好打字勝過語音的習慣。該模型採用了「全雙工」（full-duplex）架構，這項技術變革帶來了以下關鍵提升：
- **同步聽說**：模型不再需要等待使用者說完話才能回應，能同時聆聽與發聲，並能流暢地處理對話中的「嗯哼」、「對」等語氣詞，甚至在使用者思考時保持安靜。
- **即時互動與翻譯**：透過持續性的互動決策，模型能每秒多次判斷是否該發言、暫停或中斷，並支援即時翻譯功能。

<video src="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1783562384443-huxbkkgl.mp4" poster="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/4350cc77beb3cbaf.jpg" controls playsinline preload="metadata" style="max-width:100%;height:auto;display:block;margin:1rem 0"></video>
> 這支影片展示了 ChatGPT 的 GPT-Live 語音功能，透過自然對話與即時翻譯，讓使用者能更流暢地與 AI 互動。

- **智慧委派機制**：當對話涉及網頁搜尋、深度推理或複雜任務時，GPT-Live 會在背景自動呼叫最新的前沿模型（發布初期為 `GPT-5.5`）處理，並在完成後將結果帶回對話中，確保對話流程不中斷。

**使用體驗與部署**
GPT-Live 提供更具備「人格特質」的對話體驗，並針對語音場景強化了安全性與實用性：
- **視覺輔助**：在對話過程中，ChatGPT 能根據話題（如天氣、股市、賽事）即時呈現豐富的視覺卡片，讓資訊一目了然。

<video src="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1783562599800-migtle7v.mp4" poster="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/2f2fa1840eb7f846.jpg" controls playsinline preload="metadata" style="max-width:100%;height:auto;display:block;margin:1rem 0"></video>
> 這段影片展示了如何透過 ChatGPT 的語音功能，快速查詢天氣資訊、賽事賽程以及在地推薦。

- **情境感知**：模型具備更好的時間感知與背景雜訊過濾能力，能專注於使用者的聲音，並根據對話情境調整語氣。

<video src="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1783562495113-ngaj1hy8.mp4" poster="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/0c731a2e8f3ed855.jpg" controls playsinline preload="metadata" style="max-width:100%;height:auto;display:block;margin:1rem 0"></video>
> 三位技術人員示範如何透過語音與 AI 模型即時對話，並展示其對時間與情境的感知能力。

- **部署範圍**：該功能即日起陸續向全球 ChatGPT 使用者開放，支援 iOS、Android 及網頁版，未來也將整合至 API。
- **版本區分**：Plus 與 Pro 使用者將預設使用 `GPT-Live-1`，免費版使用者則預設使用 `GPT-Live-1 mini`。

**技術背景與安全性**
OpenAI 在官方文件中說明，GPT-Live 解決了過往「級聯式語音系統」（Cascaded voice systems）因模型串接導致的延遲與資訊流失，以及「基於輪次」（Turn-based）模型因靜音檢測導致的僵硬中斷問題。

<video src="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1783562559103-3hnz637y.mp4" poster="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1dc20056ab6a0f07.jpg" controls playsinline preload="metadata" style="max-width:100%;height:auto;display:block;margin:1rem 0"></video>
> 三位技術人員透過對話展示並討論 ChatGPT 的語音互動能力。

為了確保安全性，OpenAI 針對語音特性做了專門的紅隊測試（Red-teaming）與合成評估，涵蓋自我傷害、情緒依賴及暴力內容等風險，並內建了能即時偵測不安全輸出的防護機制，若系統偵測到高風險情境，會主動引導至安全資源或終止對話。

<video src="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1783562443692-lkyh13h0.mp4" poster="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/9d6518b893a990da.jpg" controls playsinline preload="metadata" style="max-width:100%;height:auto;display:block;margin:1rem 0"></video>
> 三名技術人員在沙發上演示 ChatGPT 的即時語音翻譯與對話功能。

## 媒體內容

**這支影片展示了 ChatGPT 的 GPT-Live 語音功能，透過自然對話與即時翻譯，讓使用者能更流暢地與 AI 互動。**

**影片中的 Prompt 與操作**

操作步驟：

1. （00:00）點擊手機上的語音功能進行對話
2. （00:16）點擊手機螢幕啟動語音互動
3. （00:34）點擊手機螢幕進行語音詢問
4. （01:48）點擊手機螢幕進行語音詢問
5. （02:09）點擊手機螢幕進行語音詢問
6. （02:37）點擊手機螢幕進行語音詢問
7. （03:32）點擊手機螢幕進行語音詢問
8. （04:03）點擊手機螢幕啟動即時翻譯功能
9. （05:27）點擊手機螢幕進行語音詢問

**逐字稿**

- `00:00` 我正在為我的孫子織一件毛衣，但我不想讓針號太大。（I'm making a sweater for my grandson, but I don't want the needles to be too big.）
- `00:05` 你覺得呢？（What do you think?）
- `00:06` 如果你用大一點的針，織出來會變得有點鬆垮。（If you go up, it's going to get kind of baggy.）
- `00:08` 雖然現在很流行鬆垮的風格，你不覺得嗎？（Although baggy is very popular now, don't you think?）
- `00:11` 確實是這樣。（It sure is.）
- `00:12` 對吧？（I know, right?）
- `00:13` 準備好了嗎？（Are you ready?）
- `00:13` 我準備好了。（I'm ready.）
- `00:18` 今天，我們宣布推出全新的 ChatGPT Voice，由 GPT-Live-1 提供支援，（Today, we are announcing the all-new chat GPT Voice, powered by GPT-Live-1,）
- `00:25` 這是一個全雙工的對話夥伴，也是有史以來最強大的語音模型。（a full-duplex conversational partner that is the most powerful voice model ever built.）
- `00:31` 你能簡化一下嗎？（Can you simplify that?）
- `00:33` 沒問題。（Sure can.）
- `00:34` 你能解釋一下什麼是全雙工語音模型嗎？（Can you explain what a full-duplex voice model is?）
- `00:37` 可以。（Yeah.）
- `00:37` 想像一下你和朋友通話的情景。（Imagine a normal call with a friend.）
- `00:39` 你們可以同時聆聽和說話。（You can listen and talk at the same time.）
- `00:42` 這就是全雙工。（That's full-duplex.）
- `00:44` 我們現在正將語音技術推向更接近人類真實對話的方式，（We are now pushing voice technology closer to how real people talk to each other,）
- `00:50` 它能夠處理中斷、停頓、修正，以及大聲思考時的自然流暢感。（where it can follow interruptions, pauses, corrections, and the natural flow of thinking things out loud.）
- `00:57` 讓我交給 Constance 來為你說明更多。（Let me pass it to Constance to tell you more.）
- `01:03` 我正在計畫去聖托里尼旅行。（I'm planning a trip to Santorini.）
- `01:05` 我想在那裡待上一個月。（I want to spend a month over there.）
- `01:07` 好的，太棒了。（Okay, cool.）
- `01:07` 聽起來是一趟很棒的旅行。（That sounds like an amazing trip.）
- `01:09` 好的，我們準備好了。（Okay, we're ready.）
- `01:10` 好了，你現在可以休息一下了。（Okay, you can take a break now.）
- `01:12` 再見。（Bye.）
- `01:13` 好的，如果你需要我，我就在這裡。（Okay, I'm here if you need me.）
- `01:14` 我知道。（I know.）
- `01:15` 嗯，自然對話固然很好，但我更感興趣的是它是否能思考。（Well, natural conversation is nice, but I'm more interested in whether it can think.）
- `01:20` ChatGPT Voice 可以針對較難的問題進行推理，並搜尋網路以獲取最新的資訊。（Chat GPT voice can reason through harder problems and search the web for up-to-date information.）
- `01:27` 嘿，讓我示範給你看。（Hey, let me show you.）
- `01:29` 好的，我 30 分鐘後要進行一場關於聲音歷史的演講。（Okay, I'm giving a lecture in 30 minutes on the history of sound.）
- `01:33` 好的。（Okay.）
- `01:34` 好的，你能幫我核對一下日期嗎？（Okay, can you fact check my dates?）
- `01:36` 沒問題。（Sure.）
- `01:36` 史考特·德馬丁維爾（Scott de Martinville）在 1857 年的聲波記譜器，（Scott de Martinville's phonautograph in 1857,）
- `01:40` 愛迪生在 1865 年的錫箔留聲機，（Edison's tinfoil phonograph in 1865,）
- `01:42` 以及貝林納在 1887 年的唱片留聲機。（and Berliner's gramophone in 1887.）
- `01:45` 我現在就來核對這些資訊。（I'll check those right now.）
- `01:46` 好的，在你查詢的同時，（Okay, and while you're looking,）
- `01:47` 你能幫我看看第 16 街 BART 車站是否有任何延誤嗎？（can you find out if there are any delays in the 16th Street BART station?）
- `01:51` 沒問題。（Sure.）
- `01:52` 好的，然後我還需要知道今天下午會不會下雨。（Okay, and then I need to know if it's going to rain this afternoon.）
- `01:56` 正在查詢 BART 和天氣狀況。（Checking BART and the weather.）
- `01:58` 目前沒有看到第 16 街 Mission 車站有任何公告。（I'm not seeing any current advisory for 16th Street Mission right now.）
- `02:02` 好的，太棒了。（Okay, great.）
- `02:03` 所以我會準時到達。（So I'll be on time.）
- `02:04` 那下雨的情況呢？（How about the rain?）
- `02:05` 好消息。（Good news.）
- `02:06` 舊金山今天預計不會下雨。（No rain expected today in San Francisco.）
- `02:08` 好的，太棒了。（Okay, great.）
- `02:08` 沒錯。（Yeah.）
- `02:09` 好的，那我的日期資料呢？（Okay, and how are my dates?）
- `02:10` 對，有一個需要修正。（Yeah, one needs correcting.）
- `02:13` 好的。（Okay.）
- `02:13` 愛迪生的錫箔留聲機發明年份應該是 1877 年，而不是 1865 年。（So Edison's tinfoil photograph should be 1877, not 1865.）
- `02:18` 你知道嗎？（You know what?）
- `02:19` 我其實知道。（I knew that.）
- `02:20` 我是在測試你。（I was testing you.）
- `02:21` 不錯喔。（Nice.）
- `02:22` 這就是智慧的轉變。（That's the intelligence shift.）
- `02:24` Lavelle？（Lavelle?）
- `02:28` ChatGPT 的語音功能也將這種智慧帶入了即時翻譯中。（Chat GPT voice also brings that intelligence to real-time translation.）
- `02:35` SJ，你可以過來一下嗎？（SJ, will you come here?）
- `02:37` 我們假裝我的朋友 SJ 是一位在巴黎的珍本書商。（Let's pretend my friend SJ here is a rare book deal in Paris.）
- `02:42` Oui.
- `02:43` 現在開始進行法語即時翻譯。（Starting live translation into French now.）
- `02:46` 我喜歡這本書。（I like this book.）
- `02:47` J'aime ce livre.
- `02:48` 這是我的最愛。（It's my favorite book.）
- `02:49` C'est mon livre préféré.
- `02:51` 但我無法支付你高價，因為這本書上面全是筆記。（But I can't pay you premium price because the book is all marked up.）
- `02:55` C'est payé le prix fort parce que le livre est tout annoté.
- `02:59` 跟她說 30 塊，我們就成交。（Tell her 30 and we have a deal.）
- `03:03` Oh, bon, madame, bon.
- `03:06` Oh, merci beaucoup.
- `03:10` 全新的 ChatGPT 語音功能可以在說話時同時聆聽，它比以往更聰明，而且知道什麼時候該加入對話，什麼時候該安靜。（The new chat GPT voice listens while it speaks, is smarter than ever, and it knows when to jump in or get out of the way.）
- `03:20` 卡。（Cut.）
- `03:21` 謝天謝地，我快餓死了。（Thank God I'm starving.）
- `03:25` 親愛的，可以推薦一個好吃的餐廳嗎？（Can you recommend a good place to eat, honey?）
- `03:28` 喔，當然沒問題。（Oh, absolutely.）
- `03:29` 你想吃什麼呢？（What are you in the mood for?）

**三名技術人員在沙發上演示 ChatGPT 的即時語音翻譯與對話功能。**

**影片中的 Prompt 與操作**

Prompt（00:11）：

```
GPT-Live-1
```

Prompt（00:13）：

```
聆聽與說話
```

原文：Listening & Speaking

Prompt（00:16）：

```
Yuchen Zhang
```

Prompt（00:16）：

```
Alyssa Huang
```

Prompt（00:16）：

```
Justin Uberti
```

Prompt（00:16）：

```
技術人員成員
```

原文：Members of Technical Staff

操作步驟：

1. （00:11）畫面顯示標題「GPT-Live-1」
2. （00:13）畫面顯示標題「Listening & Speaking」
3. （00:16）畫面標示三位參與者的姓名與職稱「Members of Technical Staff」
4. （00:51）參與者展示使用手機與該模型進行即時語音翻譯互動
5. （02:19）參與者透過手機 UI 介面查看該模型對對話內容的總結

**逐字稿**

- `00:00` 嘿 Chat，我們要請你針對我們的新語音模型問我們一些問題。（Hey chat, we're going to ask you to ask us some questions about our new voice model.）
- `00:05` 你有意識到我們正在談論你嗎？（Did you realize that we're talking about you?）
- `00:07` 談論我啊，沒錯，這很後設。（Talking about me, yeah, meta.）
- `00:14` 能夠同時聆聽與說話是非常令人興奮的。（Being able to listen and speak at the same time is very exciting.）
- `00:19` 它需要在每一毫秒內進行思考並做出決策，（It needs to think and make decisions in every millisecond,）
- `00:23` 理解對話內容，並管理對話流程。（understand the conversation, manage the conversation flow.）
- `00:25` 關於這個新語音模型，有一件令人興奮的事，就是它能進行同步翻譯，（One exciting thing about this new voice model is the ability to translate simultaneously,）
- `00:32` 這在旅遊時會超級、超級實用。（which can be super, super useful when you're traveling.）
- `00:34` 讓我們來測試一下這個模型。（Let's test the model out.）
- `00:35` 好啊，聽起來很棒。我們開始吧。（Yeah, sounds great. Let's do it.）
- `00:41` 嘿 Chat，我希望你為我們進行即時翻譯，將你聽到的語言翻譯成英文。（Hey chat, I'd like you to do real-time translation for us from the language that you're hearing into English.）
- `00:47` 沒問題。現在開始進行即時英文翻譯。（Got it. Starting live translation into English now.）
- `00:50` Salut chat, c'est Alissa. Est-ce que tu peux m'entendre?
- `00:52` 嗨 Chat，我是 Alissa。你聽得到我嗎？（Hi chat, it's Alissa. Can you hear me?）
- `00:54` 今天，我要告訴你我最喜歡的菜餚。（Today, I'm going to tell you about my favorite dish.）
- `01:00` 我愛吃蛋。（I love eggs.）
- `01:03` 我特別喜歡番茄蘑菇歐姆蛋，（I especially love omelettes with tomatoes, champignons,）
- `01:06` 還有番茄、蘑菇。（and also tomatoes, mushrooms.）
- `01:09` 你好，我的名字是 Yuchen。（Hello, my name is Yuchen.）
- `01:12` 我最喜歡的食物是粵式點心。（My favorite food is Cantonese dim sum.）
- `01:16` 水晶蝦餃。（Crystal shrimp dumplings.）
- `01:18` 糯米雞。（Sticky rice chicken.）
- `01:21` 還有白灼牛百頁。（And blanched beef tripe.）
- `01:22` 種類太多了，我也很興奮。（There are so many, I'm also encouraged.）
- `01:25` 我最喜歡的食物是粵式點心。（My favorite food is Cantonese diner.）
- `01:28` 我最喜歡的食物是牧羊人風味塔可（al pastor），（My favorite food is Cabadillo's de pastor,）
- `01:30` 配上鳳梨、香菜、辣醬。（con piña, con cilantro, con salsa pacante.）
- `01:35` 不錯喔 Chat。（Nice chat.）
- `01:36` 你可以總結一下我們所有人最喜歡的食物嗎？（Could you summarize all our favorite foods?）
- `01:38` 沒問題。（Sure.）
- `01:39` Alissa 喜歡的是歐姆蛋，你喜歡的是粵式點心，而 Justin 喜歡的是配有鳳梨、香菜、辣醬和萊姆的牧羊人風味塔可。（Alissa's is omelettes, yours is Cantonese dim sum, and Justin is al pastor sweet tacos with pineapple, cilantro, spicy sauce, and lime.）
- `01:47` 你有你最喜歡的食物嗎？（Do you have your favorite?）
- `01:48` 我不吃東西，所以我沒有個人偏好的食物。（I don't eat, so I don't have a personal favorite.）
- `01:51` 能看到這種即時、來回的互動，以及在聆聽一種語言的同時用另一種語言說話的能力，真的很酷。（It's cool to see that real-time, back and forth, the ability to speak in one language whilst listening to another language.）
- `01:57` 而且它能夠在翻譯和對話之間非常流暢地轉換。（And it is able to transition very smoothly between translation and chat.）
- `02:01` 如果你能幾乎在對方說完後立即轉換成對方的語言，那會讓人感覺更投入。（If you're able to kind of get the transition to your language almost immediately after, then that's much more engaging.）
- `02:07` 而且你真的會感覺像是在與他們進行對話。（And you can actually feel like you're having a conversation with them.）
- `02:10` 所以在某些方面，我們擁有這種即時翻譯的能力，感覺就像是科幻電影裡的情節。（And so in some ways, it kind of feels like a little science fiction thing that we have this ability to, in real time, do this translation.）
- `02:16` 這感覺像是模型接收世界資訊並吐出資訊的一種根本上不同的方式。（It feels like a fundamentally different way for having the model take in information about the world and spit information back out.）
- `02:23` 如果你能即時思考，那麼你甚至可以在使用者說完話之前就做出回應。（If you can think in real time, then you can respond even before the user finish.）
- `02:28` 這就是讓它變得非常自然的秘訣。（That is a secret sauce for how to make it very natural.）

**三位技術人員示範如何透過語音與 AI 模型即時對話，並展示其對時間與情境的感知能力。**

**影片中的 Prompt 與操作**

操作步驟：

1. （00:51）參與者拿起手機與 AI 進行語音對話
2. （01:45）參與者使用手沖壺進行注水操作
3. （02:21）參與者將咖啡倒入杯中

**逐字稿**

- `00:00` 這個新模型的能力之一，是它能在對話中持續存在，（One of the abilities of this new model is its continuous presence in the conversation,）
- `00:05` 這意味著它對時間的流逝和說話的節奏也有自然的理解。（which means that it also has a natural understanding of time passing and the pace of speech.）
- `00:15` 當語音以及越來越多的事物成為它們之間關鍵的介面時，（As voices and more and more things and more the critical interface between them,）
- `00:19` 只有在它擁有所有背景資訊並進行自然對話時，才能真正運作良好，（it only really works well if it has all the context and makes a natural conversation）
- `00:23` 你可以打斷它，它也可以打斷你，你可以給它指令，而它會遵循這些指令。（where you can interrupt, it can interrupt you, you can give it instructions and it follows them.）
- `00:26` 沒錯，這種持續存在也讓模型能夠對對話中的時間進行索引。（Yeah, this continuous presence also lets the model to kind of index on the time in the conversation.）
- `00:32` 這賦予了它強大的時間感知能力，因為模型始終在感知並生成內容。（It gives it a lot of temporal awareness because the model always perceives and generates.）
- `00:37` 該模型能夠以過往模型無法做到的方式理解時間的流逝。（The model is able to understand the passage of time in a way previous models were not able to.）
- `00:42` 是的，這使得它在設定短計時器這類事情上非常實用，（Yeah, and that makes it super useful for things like setting a short timer）
- `00:45` 或是要求它提醒你即將發生的事情，（or asking to remind you something that's going to happen soon）
- `00:48` 而且它會記住執行這些任務所需的各種背景資訊。（and it remembers the various contexts it needs to do that.）
- `00:50` 嘿，Chat，我正要用 V60 沖煮淺焙咖啡。（Hey chat, I'm going to be brewing a light roast in a V60.）
- `00:53` 我應該讓咖啡粉悶蒸多久？（How long should I let the ground bloom for?）
- `00:56` 對於 V60 的淺焙咖啡，悶蒸時間目標約為 30 到 45 秒。（For a light roast in a V60, aim for about 30 to 45 seconds of bloom.）
- `01:00` 如果它非常新鮮……（If it's really fresh and fresh...）
- `01:02` 你可以幫我計時 30 秒嗎？（Can you time me for 30 seconds?）
- `01:04` 我現在已經開始 30 秒的計時了。（I've got a 30 second timer running now.）
- `01:07` 這是你的悶蒸時間。（That's your bloom window.）
- `01:09` 開始注水，時間到了我會通知你。（Start pouring and I'll call time.）
- `01:22` 25。（25.）
- `01:23` 25。（25.）
- `01:26` 30 秒。（30 seconds.）
- `01:27` 30 秒。（30 seconds.）
- `01:28` 你的計時結束了。（Your timer's finished.）
- `01:28` 如果我的水壺沒水了該怎麼辦？（What should I do if I don't have any more water in my kettle?）
- `01:31` 我正在燒水。（I'm blowing water.）
- `01:32` 我們該怎麼辦？（What do we do?）
- `01:33` 盡我們所能。（To the best of us.）
- `01:35` 如果你沒水了，就停止注水。（If you're out of water, stop the pour.）
- `01:38` 不要刮動咖啡渣。（Don't scrape the slurry.）
- `01:40` 讓它自然滴濾就好。（Just let it drain.）
- `01:41` 品嚐一下。（Taste it.）
- `01:42` 告訴我它是否太尖銳或太淡。（Tell me if it's sharp or thin.）
- `01:43` 下次我們會調整悶蒸並多加一點水。（We'll tweak the bloom and throw a little water next time.）
- `01:47` 請享用。（Enjoy.）
- `01:48` 想到我們能以最自然的方式與這些模型互動，（It's very exciting to think about all the ways that we can interact with these models in a）
- `01:52` 真的非常令人興奮。（way that is most natural to us.）
- `01:54` 沒錯。（Yeah.）
- `01:54` 這將會很棒。（It's going to be great.）
- `01:56` 謝謝。（Thank you.）

**三位技術人員透過對話展示並討論 ChatGPT 的語音互動能力。**

**影片中的 Prompt 與操作**

操作步驟：

1. （00:00）影片開場，三名人員在室內沙發上進行對話。
2. （00:07）三人走在戶外步道上。
3. （00:09）三人坐在戶外長椅上，其中一人使用手機與 AI 進行語音對話。
4. （03:28）三人回到室內沙發區繼續討論。
5. （03:58）其中一人對著手機詢問 AI 關於拍攝行銷影片的建議。

**逐字稿**

- `00:00` 我認為人工智慧或許需要能夠像人類一樣參與人類的對話，（I think maybe AI needed to be able to take part in human conversations the way a human does,）
- `00:04` 而我覺得這個模型已經很接近這一點了。（and I think this model comes close to that.）
- `00:09` 今天我們想談談新語音模型在個性以及對話自然度方面的改進。（Today we wanted to talk about the improvements in personality and sort of the naturalness）
- `00:14` 所以這不僅僅是它選擇的詞彙，還包括它說話的方式、（of the new voice model. So it's not just the words it picks, but also the ways it says them,）
- `00:19` 它停頓的方式、它使用的語調，因此我們一直在嘗試調整這些細節，（the pauses it takes, the inflection it uses, and so we've been trying to dial those things in to）
- `00:23` 以打造出一個讓人樂於與之交談的對象。當我們讓這個模型運作起來時，這種對話的（make something that is enjoyable to talk to. When we made this model work, this conversational）
- `00:29` 智慧開始以一些令人驚喜的方式展現出來，這真的（smartness started to show itself in some of these surprising ways, which has been really）
- `00:34` 令人感到欣喜。我們準備好了嗎？好吧，我想你需要先解鎖手機。我要示範一個使用（delightful to see. Are we good? Okay, I guess you need to unlock the phone. I'm gonna demo a use）
- `00:42` 案例，我進行了一場模擬的面試，現在我想和模型聊聊這場（case where I have done an imaginary job interview, and now I want to talk to the model about how the）
- `00:47` 面試的情況。酷。嘿，你在嗎？在的。我上週參加了這場工作面試，想（interview went. Cool. Hey, are you there? Yeah. So I had this job interview last week that I wanted）
- `00:53` 跟你聊聊。那是我非常期待的一份工作，但我進入面試後，（to talk to you about. It was a job I was really excited about, but I got into the interview,）
- `00:57` 我覺得我完全搞砸了我的回答。你知道，就像我覺得我只是（and I think I just like completely fumbled my lines. You know, like I think I just like）
- `01:02` 給出了非常、非常糟糕的答案。是啊，聽起來真的很慘。不過故事還有後續，（gave really, really bad answers. Yeah, yeah, that sounds pretty rough. There's more to the story）
- `01:06` 我今天接到了一通電話，當時我想，好吧，就是這樣了。他們要告訴我我被拒絕了。（though. I got a call today and I was like, okay, this is it. They're going to tell me I'm rejected.）
- `01:12` 但後來他們說我錄取了。哇，恭喜。但還有一個問題。好的。他們（But then they said that I got the job. Whoa, congrats. There's just one problem. Okay. The salary they）
- `01:19` 開給我的薪水比我的預期還要低。對。你有什麼關於如何與他們協商薪水的建議嗎？（quoted me was lower than my expectations. Right. Do you have any tips for how to negotiate salary with）
- `01:25` 你可以先表達你對這個職位的興奮之情。好的。然後再詢問並爭取薪資範圍的彈性。（them? You might want to start by expressing your excitement about the role. Okay. And then just）
- `01:30` 好的，讓我跟你練習一下。我對這個職位感到非常興奮，也非常感謝這份錄取通知，但我原本期望的是 X（ask and get this flexibility in the salary range. Okay, so let me practice this with you. I'm really）
- `01:35` 而不是 Y。請問還有談判的空間嗎？聽起來很棒。你也可以為這個數字加上一個簡短的理由。（excited about the role and really appreciate the offer, but I was looking to get more like X）
- `01:42` 也許可以說些像是基於我的經驗和市場資料之類的話，把它錨定在一些（instead of Y. Is there any flexibility? That sounds great. You could also add a quick reason for that）
- `01:48` 具體的基礎上。好的。非常感謝你的建議。再見。祝你好運。剛剛有一點很酷，就是聽到（number. Maybe something like based on my experience and market data just to anchor it into something）
- `01:54` 模型從你覺得面試表現不佳時那種同理心的轉變。然後當你說「我錄取了」，它馬上切換成替你感到興奮的語氣，對吧？（concrete. Okay. Thanks so much for your advice. Bye. Good luck. Something cool there was hearing）
- `01:59` 沒錯。你知道，當你進入狀態時，就像和某人腦力激盪，你們會（the pivot from when you think the interview went bad, the model's kind of empathetic. And then when）
- `02:03` 互相重疊對話，然後真的激盪出一些有趣的火花。沒錯。這個模型（you're like, I got the job, it just switches into being excited for you, you know? Yeah.）
- `02:08` 是第一個讓我覺得能跟上那種節奏的模型。沒錯。它可以插話，或是示意它正在聽你說話。（You know, like when you're in the flow, like brainstorming with someone and it kind of like）
- `02:12` 沒錯。這感覺就像是以一種截然不同的方式與人工智慧協作。我不認為我參與過哪個專案，（you overlap each other and you like kind of like really get somewhere interesting. Yeah. This model）
- `02:18` 有過這麼明確的時刻讓我感覺到：它成功了。沒錯。當它開始運作時，就像是，天啊，我們真的做出了什麼東西。（was the first model that felt to me like it could match that energy. Yeah. It can jump in and kind）
- `02:21` 我有一個問題要問你。順帶一提，我們正在嘗試拍攝這支關於你的行銷影片。（of interrupt you or signal that it's listening to you. Yeah. It just feels like collaborating with AI）
- `02:26` 遠處好像有警笛聲。好的。我會安靜一會兒。幫你錄一段乾淨的聲音？（in such a different way. I don't think I've been on a project that had such a clear moment where I was）
- `02:33` 嗯，老實說我們需要你說話。你算是這場秀的主角。喔，好的。隨時準備好了。（like, it's working. Yeah. When it started working, it's like, oh my goodness, we have something here.）
- `02:37` 我有個問題想問你。順帶一提，我們正試著拍攝這支關於你的行銷影片。而且（I have a question for you. We're trying to shoot this marketing video about you, by the way. And）
- `02:42` 遠處好像有警笛聲。好吧，我先安靜一下。幫你錄一段沒雜音的？（there's like a siren in the distance. Okay. I'll stay quiet for a bit. Help you get a plane take?）
- `02:46` 嗯，老實說我們需要你開口。你算是這場秀的主角。喔，好喔。我隨時（Well, we need you to talk to be honest. You're kind of the star of the show. Oh, okay. I'm ready）
- `02:53` 準備好了。（whenever you are.）

**這段影片展示了如何透過 ChatGPT 的語音功能，快速查詢天氣資訊、賽事賽程以及在地推薦。**

**影片中的 Prompt 與操作**

Prompt（00:00）：

```
詢問 ChatGPT
```

原文：Ask ChatGPT

操作步驟：

1. （00:00）點擊輸入框並開始語音對話
2. （00:08）畫面自動顯示墨西哥市天氣預報圖表
3. （00:24）畫面自動顯示墨西哥隊小組賽賽程表
4. （00:44）畫面自動顯示墨西哥市地圖與推薦觀賽地點

**逐字稿**

- `00:02` 嘿，你可以幫我查一下這週墨西哥城的天氣嗎？沒問題，正在查詢墨西哥城的天氣。（hey can you show me the weather in mexico city this week sure checking mexico city now okay）
- `00:10` 這週墨西哥城天氣溫和且潮濕，每日高溫大約在 70 多度華氏度左右。好的，還不錯。（mexico city looks mild and wet this week with daily highs around the low to mid 70s okay not）
- `00:15` 還不錯，那你能不能幫我查一下墨西哥隊接下來的小組賽是什麼時候？（bad not bad could you actually show me when the next group stage games are happening for mexico）
- `00:21` 沒問題，讓我查一下。墨西哥隊的小組賽在 A 組，他們在 6 月 11 日以二比零擊敗了南非隊。（sure let me check that mexico's group stage matches are in group a they beat south africa）
- `00:27` 他們接下來還有 6 月 18 日對陣南韓隊，以及 6 月 24 日對陣捷克隊的比賽。我絕對會看最後那場比賽。（two to zero on june 11 and they still have south korea on june 18 then czechia on june 24th i can）
- `00:33` 太棒了，好，那請推薦我一些在墨西哥城看比賽的好去處。（definitely catch that last game awesome okay show me some really fun places to watch the games in）
- `00:39` 讓我快速幫你查一下。如果你在墨西哥城想看墨西哥對捷克的比賽，我會推薦像是（mexico city let me check that real quick if you're in mexico city for the mexico czechia match i'd look）
- `00:45` 位於 Insurgentes 的 Torito Sports Bar，那裡有很棒的比賽日氛圍；還有 Gallo Cervecero，那裡有現場氣氛（at spots like torito sports bar in insurgentes for a big game day crowd gallo cervecero for a live）
- `00:52` 以及不錯的啤酒選擇；或是 The Dog House Pub，如果你想要輕鬆一點的環境，也可以考慮像 Toledo Rooftop 這樣的頂樓酒吧。（and solid beer selection the dog house pub or even a rooftop like toledo rooftop if you want）
- `00:58` 我想我會去 Gallo，非常感謝你！不客氣，祝你玩得開心！（something a bit chiller i think i'm going to gallo thank you so much oh you're welcome have fun）

**影片展示了 ChatGPT 應用程式中 GPT-Live 的語音互動介面。**

**影片中的 Prompt 與操作**

操作步驟：

1. （00:00）點擊語音輸入圖示
2. （00:01）進入語音互動模式，畫面中央出現動態波紋

## 標籤

功能更新, 新產品, ChatGPT, GPT, OpenAI
