# Matthew Berman 訪談 Tibo：Astra 與新模型如何改變開發者工作流程

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

> 原作者：Matthew Berman (@MatthewBerman) · 策展與摘要：EasyVibeCoding · 平台：X (Twitter) · 熱度：🔥🔥🔥🔥🔥 · 日期：2026-08-25

> 原始來源：https://x.com/MatthewBerman/status/2091959711423996249

## 證據與延伸閱讀

- [Matthew Berman 訪談 Tibo：Astra 與新模型如何改變開發者工作流程。](https://x.com/MatthewBerman/status/2091959711423996249)
- [社群稱Tibo為reset lord，Matthew Berman回應澄清](https://x.com/MatthewBerman/status/2091966347542737098)
- [Chris主張AI環境影響微小且能帶來效益](https://x.com/ChrisGPT/status/2092025813809959409)
- [Matthew Berman澄清提問動機並認為環境問題被誇大](https://x.com/MatthewBerman/status/2092034279907987868)

## 中文摘要

Matthew Berman 訪談 Tibo：Astra 與新模型如何改變開發者工作流程。

訪談也觸及使用限制反覆重設、自我遞迴改進、AI 風險與環境影響，呈現產業加速發展下的期待與爭議。

**訪談重點** 影片章節顯示，Tibo 分享了在 Google DeepMind 的學習（0:45），並說明 OpenAI 如何建立「relentless culture」（4:22）。其他主題包括：

- Astra 與新一代模型（7:23）
- AI 加速改變開發者工作流程（11:18）
- ChatGPT 與 Codex 的整合方向（14:27）
- OpenAI 與 Anthropic 的比較（20:25）
- OpenAI 為何持續重設使用限制（23:37）
- Recursive Self-Improvement（30:25）
- 導致「The Pause」的危險（32:00）
- Ultra Fast 是否會成為預設模式（34:13）
- 為何每個人都應該嘗試 AI（43:20）

**「Reset」形象** 社群回應集中在 Tibo 經常重設使用者限制的形象。Emmanuel Crown 稱這是首次看見「大重設按鈕」背後的人；其他留言則以「reset lord」和「替我拯救使用限制的人」戲稱他。Matthew Berman 回應，Tibo 在 OpenAI 實際負責的工作不只是按下重設按鈕，顯示這個網路稱號只反映其公開形象的一部分，不能代表他的完整職責。

**環境爭議** 訪談末段提到 AI 的環境影響後，Chris 詢問這裡所指的是用水、碳排放還是能源消耗，並主張 AI 對水資源與碳排放的實際貢獻，相較其他技術與產業「極其微小」。他進一步認為，AI 可能帶來環境與醫療效益，抑制 AI 反而可能造成更大的環境損害與更高的癌症死亡率。不過這是 Chris 的立場，並非訪談提供的研究結論。

**發布者立場** Matthew Berman澄清，他提問不是要表達自己對相關議題的看法，而是希望回應一般大眾對 AI 環境影響的擔憂；他並表示，自己認為外界對 AI 環境問題的描述被過度放大。這段討論反映出，AI 的環境成本仍是公共辯論的一部分，但本次對話並未提供具體排放、用水或能源數據。

<video src="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/1787644775561-vffgt7yu.mp4" poster="https://pub-75d4fe1e4e80421b9ecb1245a7ae0d1a.r2.dev/curated/bd538d7e616d4a5e.jpg" controls playsinline preload="metadata" style="max-width:100%;height:auto;display:block;margin:1rem 0"></video>

## 媒體內容

**逐字稿**

- `00:00` 我想什麼時候按下按鈕都可以，只要覺得時機對了就行。（I can press the button whenever I want, whenever it feels right.）
- `00:03` 我不太會一直關注競爭對手。（I don't tend to look at the competition that much.）
- `00:05` 我真正關注的是，我們有哪些事情能做得特別好，以及我們的價值觀是什麼（I really look at what can we do uniquely well and what are our values）
- `00:09` 還有，我們要如何朝那個方向加速到最大。（and how do we maximally accelerate towards that.）
- `00:11` 也許再過一、兩年，這些速度就會變成，也許不會是預設值，（Maybe a year or two, these speeds will become maybe, if not the default,）
- `00:16` 但會非常接近預設值。（very close to the default.）
- `00:17` 而且你看看 Luna 的成本，對吧？那真的很驚人。（And you look at the cost of Luna, right? It's phenomenal.）
- `00:20` 科技有辦法隨著時間變得非常、非常有效率。（Technology has a way to become very, very efficient over time.）
- `00:23` 我們非常重視廣泛的使用權，也在針對你從這個方向中獲得的實用性進行最佳化。（We're very focused on very broad access and we're optimizing for the utility that you get out of the direction.）
- `00:29` 我聽說現在真的有實體按鈕了。（I heard there's an actual physical button now.）
- `00:31` 沒錯，有的。我拿給你看。真的非常、非常酷。（Yes, there is. I will show it to you. It's like very, very cool.）
- `00:40` Tibo，非常感謝你來接受我的訪問。（Tibo, thank you so much for joining me.）
- `00:42` 當然。對，很高興來到這裡。（Of course. Yeah, glad to be here.）
- `00:44` 對，我真的很期待和你聊聊。（Yeah, really excited to talk to you.）
- `00:45` 我想先從你在 Google 的時期開始談。（I want to actually start with your time at Google.）
- `00:48` 你當時在 DeepMind 團隊，而且在 ChatGPT 出現之前，（You were on the DeepMind team and before ChatGPT,）
- `00:52` Google 有個叫做 LMChat 的東西（Google had something called LMChat）
- `00:55` 你當時發文說，Google 太緊張了，不敢發布它。（and you had tweeted, Google was too nervous to release it.）
- `00:59` DeepMind 被阻止推出可能顛覆 Google 的產品。（DeepMind was blocked from shipping products that could disrupt Google.）
- `01:02` 我一直很常想到這件事。（And I think about that a lot.）
- `01:03` 那時候你在開發這些產品時，心裡在想什麼？（What were you thinking at that time while you were working on these products）
- `01:07` 畢竟那是在 ChatGPT 真正改變世界之前很久的事。（that was, you know, well before ChatGPT really changed the world?）
- `01:11` 對，那是一段非常令人興奮的時期。（Yeah, it was a very exciting time.）
- `01:14` DeepMind 是個非常有創意的地方。（So DeepMind was a very creative place.）
- `01:16` 我主要專注在自己的專長，也就是基礎架構，以及能加速研究的產品。（I was mostly focused on my specialty was infrastructure and products to accelerate research.）
- `01:23` 所以當然有一個團隊在做大型語言模型，以及它的擴展。（And so there was obviously a group working on language models and scaling that.）
- `01:30` 而且他們已經取得了相當不錯的成果。（And then they had gotten pretty good results.）
- `01:33` 接著，自然就會開始想，欸，你知道，（And then it was like a natural thing to think about, hey, you know,）
- `01:36` 能不能把它變成某種可以和你聊天的東西，（can you turn that into something that, you know, you can chat to）
- `01:39` 而且你知道，可以拿來做各種事情。（and, you know, can use, you know, for various things.）
- `01:43` 所以很自然地，（And so then naturally,）
- `01:44` 像 LMChat 這樣的想法就出現了。（like the idea of like something like LMChat sort of emerges.）
- `01:47` 接著它就成了內部工具。（And then it was internal.）
- `01:48` 後來也產生了把它做成公開工具的企圖。（And then there was sort of this ambition to make it into a publicly available tool.）
- `01:54` 這是幾年的事？（What year was this?）
- `01:56` 大概是在 ChatGPT 出現前一年。（And this was, it was like a year before ChatGPT roughly.）
- `01:59` 好。（Okay.）
- `02:00` 對。（Yeah.）
- `02:01` 然後，（And then,）
- `02:01` 不過我們當時也在開發各式各樣其他的東西，這些我就不談了，（but we were also building all sorts of other things that I'm not going to talk about,）
- `02:05` 但那是一個非常有創意的地方。（but it was a very creative place.）
- `02:07` 只是 DeepMind 當時並沒有為推出產品做好準備。（And then just DeepMind was not set up to ship product.）
- `02:13` 在這方面，OpenAI 是非常、非常不同的地方。（OpenAI is a very, very different place in that sense.）
- `02:17` 我們就是讓研究和產品團隊非常緊密地合作。（We just like research and product just collaborate super closely together.）
- `02:21` 我們一起構思。（We ideate together.）
- `02:21` 很多事情也是一起設計的。（We co-design a lot of things.）
- `02:23` 我們非常傾向於推出產品，也非常傾向於（We have a big bias to ship and also a big bias towards, you know,）
- `02:29` 讓人們能夠使用這些東西，而這點我真的很喜歡。（making things available for people, which I really love.）
- `02:32` 這大概就是吸引我來到這裡的原因：使命、團隊，（And this is sort of like what drove me here, the mission, the people,）
- `02:36` 以及人才密度。（the talent density.）
- `02:37` 我的意思是，OpenAI 真的有這麼多很棒的地方。（I mean, there's so many great things about OpenAI, really.）
- `02:39` 你當時參與 LMChat 的時候，有沒有意識到那是件很特別的事，（Did you know at the time where you were involved in LMChat that that was something special）
- `02:44` 或是它將會變成一件很特別的事？（or would become something special?）
- `02:45` 那感覺，那感覺真的很特別。（It felt, it felt very special.）
- `02:47` 就像那些模型，你知道，就有點像是第一次，你知道，（Like the models were, you know, it's like sort of like the first time, you know,）
- `02:51` 你意識到自己可以取得連貫的文字，而且，你知道，還能得到一些有幫助的東西。（you realize like you could get coherent text and, you know, something helpful.）
- `02:56` 一開始，它帶來的笑點比幫助還多。（Initially, it was more funny than helpful.）
- `02:58` 然後它逐漸變得越來越有幫助。（And then gradually it became more and more helpful.）
- `03:00` 所以你說，你常常會想到那段經歷。（So you say you think about that often.）
- `03:03` 我能理解。（And I understand that.）
- `03:05` 我想，你知道，Google 在很多方面都是自己阻礙了自己。（I think, you know, in a lot of ways, Google got in their own way.）
- `03:08` 你在那裡學到哪些教訓，後來帶到了 OpenAI？（What are some of the lessons that you learned there that you took to OpenAI?）
- `03:12` 對。（Yeah.）
- `03:13` 所以這就是我常常想到它的原因。（So this is why I think about it often.）
- `03:14` 我會從我在團隊中建立的文化，（I think about it in terms of like the culture that I have on the team,）
- `03:18` 以及 OpenAI 本身的文化來思考這件事。（the culture of OpenAI itself.）
- `03:20` 所以，哪些好的部分要保留，以及，你知道，哪些事不要做。（And so like the good parts to preserve and like, you know, what not to do.）
- `03:23` OpenAI 有一種非常由下而上的文化。（OpenAI has a very bottoms up culture.）
- `03:26` 這是一種非常重視賦能的文化。（Like it's a very empowering culture.）
- `03:28` 大家可以提出各式各樣的想法，聚在一起，然後非常、（Like people can come up with all sorts of ideas and get together and then very,）
- `03:32` 非常快速地把東西推出去。（very quickly ship something.）
- `03:33` 一般來說，新產品想法幾乎不會遇到什麼阻力，（And there's very little stop energy in general for like new product ideas,）
- `03:38` 這令人振奮，也很有趣。（which is exhilarating and fun.）
- `03:40` 而且，你知道，一切都是為了以正面的方式影響世界。（And, you know, it's all about impacting the world in positive ways.）
- `03:44` 所以，保留這一點對我來說非常重要。（And so preserving that is very important to me.）
- `03:46` 另一件同樣重要的事，是不要把一切弄得一團亂，對吧？（The other thing that is also important is like to not make it a mess, right?）
- `03:51` 你不會想要有一堆東拼西湊的功能，卻沒有整體方向，（So you don't want to have like a hodgepodge of like no features and like no overall direction）
- `03:55` 也沒有一致性。（and coherence.）
- `03:56` 所以，這需要用簡潔感來取得平衡，也要讓你以產品的（And so it's counterbalanced with the sense of simplicity and you being proud about the）
- `04:03` 品質為傲。（quality of the product.）
- `04:04` 我想 ChatGPT 的 iOS 應用程式算是市面上最棒的應用程式之一。（I think the ChatGPT iOS app is like, you know, one of the best apps out there.）
- `04:08` 所以我們想維持這一點。（And so we want to keep that.）
- `04:10` 我們投入了很多資源在讓人感到愉悅、效能、效率、（We're investing a lot in, you know, things like delight, performance, efficiency,）
- `04:14` 以及簡潔性這些事情上。（simplicity.）
- `04:15` 所以，在賦予每個人權力的同時，還是有這些整體原則，（So there's these overall principles while still empowering everyone,）
- `04:19` 讓每個人都能（everyone to like）
- `04:19` 嘗試新事物，並非常快速地推出產品。（try new things and ship very quickly.）
- `04:21` 如果要給創辦人一些關於如何培養這種文化的建議，（If you were to give advice to a founder about how to develop that kind of culture,）
- `04:27` 有哪些（like what）
- `04:28` 比較具體的要素或做法，是在 OpenAI 內部實際運作，能夠（are some of the more tangible elements or practices that occur inside OpenAI that can）
- `04:33` 給創辦人一些方向？（kind of give advice to a founder?）
- `04:36` 是的，（Yes,）
- `04:36` 我想，首先要有信念，並設法接觸使用者，根據回饋非常快速地反覆迭代，（I think having having conviction and finding a way to be to have users and iterate）
- `04:46` 同時也要願意顛覆自己。（very quickly from feedback and then also being willing to disrupt yourself.）
- `04:50` 這對創辦人來說不算那麼相關，但對像 OpenAI 這樣的（That is not as much relevant for founder, but as relevant for, you know,）
- `04:54` 公司來說，（companies like）
- `04:54` 我們一直都在提出新的研究、（OpenAI, like we come up with new research,）
- `04:57` 新的想法，而且能夠（new ideas all the time and being able to）
- `04:59` 辨識什麼時候是投入資源發展它們的適當時機，（identify when is the right moment to go and invest in them,）
- `05:02` 即使這代表（even though it means like）
- `05:03` 可能要從主要業務重新分配資源，你知道，這非常重要，（maybe reallocating resources from, you know, the main gig, it's super important,）
- `05:08` 但這（but it's）
- `05:09` 非常困難，但能做到這點非常重要。（very hard, but it's super important to be able to do that.）
- `05:11` 對。（Yeah.）
- `05:11` 我的意思是，這正是你剛才描述在 Google 發生的事情。（I mean, that's the exact thing that you were describing at Google.）
- `05:14` 他們某種程度上沒辦法做到這點。（They kind of weren't able to do that.）
- `05:16` 那很棒。（That's great.）
- `05:17` 我的意思是，公平地說，他們是有計畫的。（I mean, does that, they have a plan to be fair.）
- `05:19` 就像，你知道，這一切都是一個大計畫的一部分。（It's like, you know, it was all, it was all part of a big plan.）
- `05:22` 但對我來說，那不是，不是，不是正確的地方。（But to me, it wasn't, it wasn't, it wasn't the right place.）
- `05:27` 在 OpenAI，或任何一家公司逐漸成熟之後，（At OpenAI or any company as it matures,）
- `05:29` 要維持那種（does that become more difficult to maintain that）
- `05:33` 持續推出產品、並且願意顛覆自己的文化，會不會變得更困難？（kind of culture of shipping and willingness to disrupt yourself?）
- `05:37` 尤其是當你，你知道，（Especially when you, you know,）
- `05:38` 如果你有一頭只是在印鈔的金雞母，而你還有（if you have a cash cow just printing money and you have）
- `05:41` 另一個新東西在這裡，可能會是某種很酷又創新的事物。（this other new thing over here, that might be something cool and innovative.）
- `05:44` 我們非常、非常著眼於未來。（We are very, very forward looking.）
- `05:48` 而 AI 的未來、它最終會呈現什麼樣子，以及人類將如何從中受益，（And the future of AI and what it will all look like and how humanity benefits,）
- `05:55` 它不會（it doesn't）
- `05:56` 真的等你，也不太在乎，你知道，（really wait or doesn't really care for, you know,）
- `05:59` 你在這裡已經建立的任何事物，（whatever you have established here,）
- `06:01` 你知道，在接下來一個月或三個月內會怎樣。（you know, over the next month or three months.）
- `06:03` 所以，你知道，我認為勇於投入非常重要，並且要，你知道，（And so, you know, I think it's very important to lean in and to, you know,）
- `06:08` 只是保持開放（just be like open）
- `06:09` 地看清楚一切將走向何方，然後，你知道，（eyed about where it's all going and then, you know,）
- `06:12` 想辦法找出該如何定位自己，（figure out like how to position yourself）
- `06:14` 這樣你，你，你才能抓住那道浪潮。（so that, you know, you, you, you do catch that wave.）
- `06:18` 你知道，即使對 OpenAI 來說，也是這樣：我們（You know, even, even for OpenAI, it's like we,）
- `06:20` 訓練模型，然後才發現它們的（we train models and then we discover their）
- `06:22` 能力。（capabilities.）
- `06:23` 也就是說，基準測試無法告訴你全部的事情。（Like we don't, benchmarks don't tell you everything.）
- `06:25` 我們必須花不少時間親自操作這些模型，才逐漸意識到，（We have to play quite a bit with the models themselves to sort of like realize it's like,）
- `06:29` 喔，你知道，也許我們還沒想過要如何（oh, you know, maybe we haven't thought about, you know,）
- `06:32` 以這種（benefiting from it in like this）
- `06:33` 特定的方式從中受益，或者，喔，它原來可以做到這件事。（specific way or like, oh, it can do this.）
- `06:36` 然後你就會，喔，我的意思是，這代表我們思考產品的方式（And then you're, you're just like, oh, that, I mean, that's a shift in like,）
- `06:38` 有所轉變，像是，你知道，我們（you know, how）
- `06:39` 對產品的想法。（we think about the product.）
- `06:40` 例如，像現在，你知道，我們，我們，我們推出了全新的語音，（For example, like right now, like, you know, we, we, we launched the new voice,）
- `06:45` ChatGPT（Chattrity）
- `06:45` 語音，使用起來非常愉快。（Voice, and it's super delightful to talk to.）
- `06:48` 它非常自然。（It's very natural.）
- `06:50` 現在它也能使用工具了。（Now it's capable of tool use as well.）
- `06:52` 對。（Yeah.）
- `06:52` 而這改變了很多事情。（And that changes things.）
- `06:53` 像是我現在花更多時間直接跟它聊天。（Like now I spend a lot more time just talking to it.）
- `06:56` 我一直在做的另一件事就是口述輸入，（Another thing I do all the time is like dictation,）
- `06:58` 因為口述輸入的品質（because the quality of the dictation is like）
- `07:00` 實在是太、太好了。（so, so good.）
- `07:01` 而且這是一種更有效率的方式，作為一種，一種方法，（And it's like much more efficient as a, as a way,）
- `07:03` 而不是像在輸入 prompt。（instead of like typing the prompt.）
- `07:05` 所以早上我就坐在那裡拿著手機，然後就會說，啦、（And so in the morning, I just like, I sit there with my phone and I'm like, blah,）
- `07:08` 啦、啦。（blah, blah.）
- `07:09` 我就想說，你知道，Chattrity 還有幾件事要做，然後（I just like, you know, a couple of things to do for Chattrity and,）
- `07:12` 然後它就會（and then it just goes）
- `07:13` 直接去做，而且它可以存取我所有的工具。（and like does it, it has access to all my tools.）
- `07:15` 對。（Yeah.）
- `07:15` 嗯，而這就像是，（Um, and that is like,）
- `07:17` 在我們有了真的很好的語音模型之前，這是不可能做到的。（that was not possible before we had like really good voice models.）
- `07:20` 所以這完全改變了你對產品的想法，突然之間變成了另一回事。（And so that completely changes in suddenly how you think about the product.）
- `07:23` 對。（Yeah.）
- `07:23` 我們繼續聊聊新的模型和新的 harness。（Let's continue talking about new models, new harnesses.）
- `07:26` 嗯，幾週前，我想先從你的另一則推文開始。（Um, a few weeks ago, I'm going to start with another one of your tweets.）
- `07:29` 因為你知道，這些推文都很猛，呃，（Cause you know, these are bangers, uh,）
- `07:31` Codex 兩到三個月後看起來就會很原始。（codex will seem primitive in two to three months.）
- `07:33` 我們即將迎來另一波重大演進。（We're about to go through another major evolution.）
- `07:35` 下一代模型需要的不只是你的筆電。（The next generation of models need more than your laptop.）
- `07:38` 嗯，我們先從 harness 開始。（Um, what areas, let's start with the harness first.）
- `07:42` 隨著模型變得更好，harness 還有哪些領域很有創新空間？（What areas of the harness are still ripe for innovation?）
- `07:45` 隨著模型變得更好。（As a model gets better.）
- `07:47` 對。（Yeah.）
- `07:48` 非常多，真的非常多。（So many, so many.）
- `07:48` 嗯，我剛剛提到語音。現在有一件事是，嗯，（Um, so I talked about voice, like one thing that, um, right now,）
- `07:53` 如果你是個很進階的（if you're like a sophisticated）
- `07:55` Codex 使用者，或是，你知道，（user of, of, of codex and, you know,）
- `07:57` 任何其他程式開發 Agent 的使用者，你多少都已經習慣了（any other coding agent is you sort of have gotten used）
- `08:01` 那麼一點卡頓感，對吧？（to a little bit of the clunkiness, right?）
- `08:03` 所以，你知道，你得管理 skill 檔案，而且，你知道，（So, you know, you have to manage skill files and, you know,）
- `08:06` 這算是一種（this is like a way to sort of like）
- `08:08` 教它一些東西的方法，但同時，（teach it stuff, but it's also,）
- `08:09` 我想很多人也發現，這有點（I think a lot of people have realized it's kind of like）
- `08:11` 難以長期維護。（hard to maintain over time.）
- `08:13` 嗯，記憶算是一回事，但它不一定總是記得，呃，（Um, the memory is sort of like a thing, but it doesn't always remember, uh,）
- `08:19` 所有事情。（everything.）
- `08:20` 比如說，如果你有 subagent，（Like if you, if you have sub agents,）
- `08:22` 你就得在意 subagent，而它就像是（you have to care about sub agents and it's like sort）
- `08:24` 慢慢建構出一個小型網路。（of like constructs a little network.）
- `08:26` 當你和它互動時，那種整體感，就像是被拆成了各個部分，嗯，（It's like the illusion kind of gets broken into various parts, um,）
- `08:30` 消失了。（when you interact with it.）
- `08:32` 但你真正想要的，其實只是某個能深刻理解你的東西，（And really what you want is just something that deeply understands you,）
- `08:36` 理解你的（understands your）
- `08:37` 目標、理解你的日常，理解你知道的，（goals, understands your day to day, understands what, you know,）
- `08:39` 也理解你的團隊正在做什麼。（your team is up to as well.）
- `08:41` 然後理想上，它會做出反應，也會主動出擊，呃，（And then optimally sort of like reacts and also is proactive, uh,）
- `08:45` 只是在（and just helps you in）
- `08:46` 你的日常中幫助你，而且不會破壞那種整體感，對吧？（your day to day and doesn't break that illusion, right?）
- `08:49` 就像那是你擁有的完美小夥伴，呃，（Of like that's this perfect little partner, uh, that you have.）
- `08:53` 所以這就是我們正在努力實現的方向。（And so that's what we're, uh, working towards.）
- `08:55` 還有一件事是，你知道，當你擁有非常、（It's another thing that, you know, you realize when you have very,）
- `08:58` 非常強大的模型時，你會發現（very powerful models is）
- `09:00` 你的筆電本身就會變成一種限制。（that your laptop kind of becomes a constraint in and by itself.）
- `09:03` 嗯，你知道，你能在筆電上完成的工作量，（Um, you know, the amount of work that you can do on a laptop,）
- `09:06` 它原本就是為人類設計的，（it was designed for humans,）
- `09:08` 對吧？（right?）
- `09:08` 所以它大致上是按照這樣的需求設計的：能夠承受你知道，（So it's designed roughly to be able to absorb the amount of work that, you know,）
- `09:13` 你能（you can）
- `09:13` 產出的量，或是你打字的速度、思考的速度，你知道，（produce or, you know, how fast you can type and how fast you can think, you know,）
- `09:17` 以及你需要開啟多少個（how many）
- `09:17` 應用程式。（applications you need open.）
- `09:18` 而這些全都是人類的限制。（And all these things are human constraints.）
- `09:21` 嗯，模型並沒有相同的限制。（Um, the model doesn't have the same constraints.）
- `09:23` 例如，模型可以，你知道，處理（The model can, you know, for example, handle, you know,）
- `09:26` 同時開啟一百個應用程式，（a hundred applications opened at the）
- `09:27` 完全沒問題，你知道，或許未來會是這樣。（same time, perfectly fine, you know, maybe in the future.）
- `09:30` 所以就資源存取而言，很明顯地，你知道，（And so in terms of access to resources, it's very clear that, you know,）
- `09:34` 未來的模型（models of the future）
- `09:35` 需要存取的資源會超過你筆電上的資源。（will need access to more than the resource of your laptop.）
- `09:38` 我的意思是，我猜你說的是雲端 Agent，然後，（Do you, I mean, I, I'm guessing you're talking about cloud agents and,）
- `09:42` 突然之間，（and all of a sudden,）
- `09:43` 比如說，當你有像超高速這種功能時，（like, you know, when you have things like ultra fast,）
- `09:46` 我們等一下會談到，（which we're going to talk about）
- `09:48` 當你的 token 速度是十倍，我想應該是十四倍，（in a little bit, when you have token speeds that are 10, I think 14 is the,）
- `09:52` 官方公布的（the stated）
- `09:53` 數字，比高速還要快十四倍，嗯，這個頻寬。（number, 14 times faster than what fast is, um, the, the bandwidth.）
- `09:58` 頻寬改變了，抱歉，頻寬限制改變了，CPU 現在就變成頻寬，（the bandwidth changes or sorry the bandwidth constraint changes uh the cpu now becomes the）
- `10:04` 真的，工具呼叫、網路工具呼叫，以及堆疊中的任何額外負擔，都會變成（bandwidth like literally tool calls network tool call any kind of overhead in the stack becomes the）
- `10:10` 限制因素。但你知道，你也可以透過同時執行多件事來補償，（the delimiting factor but then you know you can compensate by doing multiple things uh concurrently）
- `10:18` 所以你可以想像，例如一個 Agent 負責探索，（as well and so you can you can think about you know having like maybe you know exploring on one）
- `10:25` 同時撰寫測試、進行編譯，或測試新的假設，全部一次完成。（and writing tests as well compiling uh you know testing a new hypothesis like all at once）
- `10:33` 這樣一來，你就不是，你是，你是在移動瓶頸，因為你知道，（and so then you're not then you you're you're shifting the bottleneck around because you know）
- `10:38` 你能夠同時做更多事，然後模型也能以非常有效率、非常快速的方式，（you're able to do more uh concurrently and then you know the model can like sort of like think）
- `10:42` 用目前的 token 速度，我發現自己會平行啟動十到十五個 Agent，（very efficiently and very quickly through it with current token speeds i find myself kicking off 10）
- `10:48` 而這對我來說會造成相當大的認知負擔，因為要進行這種情境切換，（15 agents in parallel and that becomes a pretty significant cognitive overhead for me to do that）
- `10:54` 而且一直都是這樣，因為你啟動任務後，通常要等三十到四十五分鐘，（context switching and just constantly because you're kicking it off and you can expect 30 45）
- `10:59` 任務才會回傳。現在有了超高速，這個工作流程會大幅改變，（minutes before my task comes back now with ultra fast speed that workflow changes significantly and）
- `11:07` 我不認為自己還能使用十到十五個 Agent，而這或許是件好事，可能一次只用（i don't think i would be able to have 10 or 15 agents and that might be a good thing maybe it's）
- `11:11` 三、四個。你覺得單人開發者的工作流程，長期下來會如何改變？（three or four at a time how do you see the the workflow of a solo developer changing over time）
- `11:16` 對，所以我認為，管理你的注意力，以及更重視你的注意力，（yeah so i think managing your attention and being much more you know friendly to your attention is）
- `11:25` 是我們非常在意的事情。畢竟，我們是為人類打造產品，想要打造（something that we care a lot about like after all like we're trying to build for humans we're trying）
- `11:29` 最能賦予人類力量的技術，而這需要圍繞著（to be like the build the technology that's the most empowering for humans and that requires building）
- `11:35` 你多工處理的能力來建構。你想怎麼管理自己的注意力？你希望（around you know your ability to multitask and you know how do you want to manage your attention and do）
- `11:40` 現在就把某件事呈現出來，還是三十分鐘後再呈現會比較好？（you want something brought up now or is it better to bring it up in 30 minutes）
- `11:45` 然後，當你把超高速和語音結合起來時，突然之間你會覺得，（and then when you have ultra fast speeds combined you know maybe with voice is like suddenly you're）
- `11:51` 好，這個東西的運作速度跟你一樣快，甚至比你更快，所以你（like okay you know like this thing can operate at the same speed if not faster than you and so you）
- `11:56` 可以持續投入其中，可以構思點子，可以看到原型，你知道，也可以（stay in the flow you get to ideate you get to see prototypes you know like you get to build little）
- `12:01` 即時建立小型報告，而這種感覺真的非常棒。嗯，你突然會覺得，（reports like in real time and that you know sort of like that just feels really good um suddenly you're）
- `12:07` 對啊，這就像我以前用十個 Agent 進行多工處理一樣，只是我真的不想（like oh yeah it's like what i was doing before multitasking like 10 agents it's like i don't want to）
- `12:12` 再回到那種方式了。對，所以我們想帶來的是一種非常自然的體驗，（really go back to that yeah and so we're trying to bring that sort of experience that is just really）
- `12:18` 同時也讓人感覺這項技術是為你打造的，而不是你必須適應技術，是技術來適應你。（natural but also feels built for you and where you don't have to adapt the technology adapts to you）
- `12:23` 所以，過去幾個月裡，已經有不少 Agentic 程式開發技術被提出討論。（so there's been a number of i guess agentic coding techniques discussed over the last few months）
- `12:28` 迴圈以前很流行，現在也還是很流行。我最近聽到有人在談圖，這些是不是都只是為了讓（loops was popular still is popular now i'm hearing about graphs are are these all techniques to just）
- `12:35` 單人開發者能夠管理自己的注意力，或是讓注意力更集中，就像你說的，我很喜歡這個說法。（allow the solo developer to manage or be friendly to their to their attention as you said i like that term）
- `12:43` 對，所以我會把問題想成兩個不同的類別。第一個是打造最好的（yeah so i i think about two different categories of problems like the first one is building the very）
- `12:51` 個人 AGI，或是能和你一起工作、主動提出重要新想法的個人 Agent，（best personal agi or the personal agent that will be in the flow with you proactive raise important new）
- `13:00` 而且一旦找到某些東西，就能非常非常有效率地完成你想做的事。（ideas uh when it can find some be very very efficient and doing exactly what you want）
- `13:06` 不管這是技術問題，還是比較像研究或提供建議的事情，其實都沒差。（it doesn't matter whether it's a technical problem or you know it's just more like research or advice）
- `13:13` 它什麼都能做，而且會高度高度地針對你的需求量身打造。這是一件非常重要的事。（like you can do it all and it's like super super tailored to you this is like a very important thing）
- `13:18` 而且這就像是深深根植於你身為一個人、身為一個（and it's like deeply rooted in like you know the understanding of you as a human you as like an）
- `13:24` 獨一無二的個體所具備的理解，這是其中一類問題，也就是我們正全力推進的方向（individual that is unique that's one category of problem like we're pushing super hard on that）
- `13:28` 另一類問題則是全面自動化，也就是你更多是在建構（the other category of problems like full-on automation um you know where you're more building）
- `13:34` 能夠處理非常複雜流程的智慧系統，也許是某些（intelligent systems that can take care of like a very complex process you know maybe something that）
- `13:39` 原本確實需要智慧，而且看起來非常複雜的事情，例如（did require you know does require intelligence and seems like very complex for example you know）
- `13:45` 查看正式環境的日誌，並自動進行效能最佳化（going and looking at production logs and automatically doing performance optimizations）
- `13:50` 我們正在處理回歸問題，並自動修補這些問題；在網路安全領域，我們也看到同樣的情況。（we're looking at regressions and automatically patching them in cyber security we're seeing this as well）
- `13:55` 也就是說，你有某個東西，像是掃描器，找出一個漏洞。（where it's like you have you know something you have a scanner that comes up with a vulnerability）
- `14:01` 你能不能自動修補它，把這個故障持續開放的時間縮短，讓它（like can you automatically patch it and reduce the window um where you have that open failure to like）
- `14:07` 幾乎降到零，而且這一切都不需要人在迴圈中，也就是不需要人介入，或是（almost zero and that's all of that with without a human in those loops without a human in the loop or）
- `14:12` 只需要非常少的人為介入，例如你只需要核准高風險的動作，就像是（like you know very very minimal where you know you only need to approve a high risk action and it's like）
- `14:18` 大部分都是自動化系統，但這也沒那麼重要，你不一定需要（mostly an automated system but it's also not that much you know it's not as important for you to be in）
- `14:25` 直接控制它。好，然後我想稍微換個話題。你知道，ChatGPT 和（direct control of it okay um and then so i want to slightly change topics and you know chat gpt and）
- `14:34` Codex 過去幾個月一直在走向整合。對，所以我想先問問你，這進展得如何？（codex have been on this merge path yeah over the last few months so i i guess first i just wanted to）
- `14:40` 在內部感覺如何？你從客戶那裡得到的回饋是什麼？（ask you how's that been going like how does it feel internally what's the feedback you've been）
- `14:43` 這確實帶來了很大的助益。起初我們得到的回饋是：（getting from your customers um it's it's really been a boon uh so the feedback we had initially was）
- `14:51` 為什麼？為什麼？為什麼要把它們合併？你真的非這麼做不可嗎？結果就像是，好吧，（like why why why do you merge them it's like you know do you really have to do it and it's like well）
- `14:55` 未來的模型希望我們把它們合併，所以，我們就會這麼做。（the the future the our future models want us to be merged so um you know we're just going to do it）
- `15:03` 因為在打造這個非常個人化、能力超強的（because it is the simple and proper thing to do where we're building this very personal super capable）
- `15:09` Agent 時，這是簡單又恰當的做法，它可以用各種方式幫助你。這會是同一套、也會是相同的技術（agent that can help you in all sorts of ways this is the same this is going to be the same technology）
- `15:14` 在幕後運作。它是同一個 harness，也是我們思考這件事的相同方式，就像是高度（under the hood um it's the same harness it's the same way that we think about it it's like highly）
- `15:19` 多模態的，你知道，以語音優先、效率超高，而且不管你是不是想要寫程式（multimodal you know voice first uh super efficient and it doesn't matter if you're trying to code or）
- `15:25` ，這個 Agent 都有能力處理一切，而且它在這方面的效率是最高、最有效率的（not like this this agent is capable of it all and it's like the high it's like the most efficient at）
- `15:30` 。至於你想要的介面，它應該要配合你的需求自行調整，而不是讓你（it and then the interface that you want is like it should tailor itself to your needs if you shouldn't）
- `15:36` 自己決定：「我是程式設計師，所以我要程式設計師介面」，或是「我不懂技術，所以我要（decide like you know i'm a coder i want a coder interface or like i'm not technical i want a）
- `15:41` 非技術介面」。人其實是分布在一個光譜上的，你知道，我們會提出（non-technical interface it's like there's a spectrum of people like you know we come up with）
- `15:44` 軟體工程師、設計師之類的標籤，但你知道，這些都只是人類創造出來的概念，用來（labels of like a software engineer a designer like you know these are just human concepts that）
- `15:48` 處理抽象事物，因為現實太複雜，我們無法直接應付（we have invented to deal with abstractions because the reality is too complex for us to handle）
- `15:53` 。但每個人就是獨立的個體，都有自己的特質，會落在這個光譜上的某個位置，而且（but if individuals are like they're individual they have their own there is somewhere on the spectrum and）
- `15:58` 所以我們想打造的是能為每個人調整的完美介面。不管你（so we're trying to build the perfect interface that adapts for everyone it doesn't matter if you're）
- `16:04` 懂不懂技術，它都會根據你獨特的個人特質來調整，所以這就是為什麼（technical or not it's just like it adapts like based on your specific individuality so that's why）
- `16:09` 我們著手做了這件事。但這是否意味著，它最後不可避免地會變成單一的（we went and we did this but it does that mean inevitably it's going to end up with a singular）
- `16:15` 介面，不再需要用下拉選單在不同產品之間選擇？想到我媽媽可能會（interface no drop down selecting between products and it's kind of wild to think that my mom might）
- `16:21` 使用和我完全一樣的介面，這件事其實滿瘋狂的。當然，它會根據我的需求自訂，也許我在做更複雜的工作時會需要更多（use the same exact interface as me and then obviously it'll customize to my needs maybe i'll need more）
- `16:27` 資訊。但對你來說，最終會是什麼樣子？沒錯（information if i'm doing more sophisticated work uh but like what is the end state for you that's right）
- `16:33` ，就是同一件事。你和你媽媽都會使用同一個東西，它會成為（it's it's the same thing um so you and your mom will you know use the same thing uh it will be）
- `16:39` 你個人的 AGI。你們會有非常不同的任務，以及從中獲得的不同用途（your personal agi you will have very different kinds of tasks and utility that you get from it）
- `16:46` 。你會把它連接到生活中的不同工具，帶入不同的想法和不同的需求（you will connect it to different tools in your life you will bring different ideas different needs）
- `16:51` ，然後它會持續自行調整，最大化地為你提供幫助。好，而且它也會（and then it will continue to tailor itself to maximally benefit you okay uh and it will you know）
- `16:57` 和你的朋友以及其他所有人一起這麼做。所以我想回到你剛才提到的一件事，你用了（do so with your friends and with everyone else so i i want to go back to something you said you used）
- `17:02` 幾次「幻覺」這個詞，來描述那種端到端的最終狀態。對一般使用者來說，那個完美的幻覺究竟是什麼？（the word illusion a couple times in that kind of end state what is that that perfect illusion for）
- `17:07` 如果你想像幾年後的我們，那麼 AI 和人類之間的（the the typical user like what like if you can envision us a few years from now what is the）
- `17:14` 互動會是什麼樣子？對我來說，它是某種非常非常（interaction between ai and a human look like yeah it's um to me it's something that is very very）
- `17:21` 量身打造給人類，嗯，而這就是為什麼大型語言模型也能成功，像是——（tailored to to to humans um and this this is why large language models are also successes like it's）
- `17:30` 這很自然，語言本來就是人類的概念，對吧？所以，你知道，我們——（it's it's natural language natural language it's like it's a human concept right uh so you know we're）
- `17:36` 習慣彼此交談。比如說，如果你明天寫信給我，我就能讀懂——（used to speaking to each other like you know if you write me a letter tomorrow i'll be able to read）
- `17:40` 嗯，你知道，就像我們現在已經相當了解彼此了，所以，我就能夠——（it um you know it's like we we know each other quite a bit now so uh you know it's like i will be able）
- `17:46` 稍微解讀信裡的一些情緒，或許也能讀出字句背後的一些細微差異——（to sort of decipher like a little bit of the emotion or you know maybe a little bit of the nuance behind）
- `17:50` 如果你寫信給我的話。嗯，而這一切都深深根植於人性，所以我們——（the letter if you wrote me a letter um and all of that is is deeply human so the technology that）
- `17:57` 正在打造的技術，都是以人性為根基，也根植於人類溝通——（we're building is you know rooted in in humanity and rooted in you know the way that humans communicate）
- `18:04` 和完成事情的方式。嗯，不應該出現這種情況：你說「喔，你——（and get things done um and there shouldn't really be a thing where you know you're like oh you）
- `18:09` 誤會我了」，因為你沒能完全解讀我語氣中的細微差異，或是你——（misunderstood me because you know you didn't quite decipher the nuance in you know my tone or you）
- `18:15` 沒完全理解文字，沒弄懂我的意思。這就是我們——（didn't quite understand the text you know how i meant it it's like that's that's um that's something）
- `18:22` 努力想避免的事。所以，我們非常努力，不是要讓你去適應，而是讓這項——（that we're trying to avoid and so we're trying to very much to not have you adapt but have the）
- `18:28` 技術像我一樣，完美地成為——嗯——成為——如同人類——（technology just like me perfectly sort of um created to um to be like a natural extension）
- `18:37` 在這個世界上原本行動方式的自然延伸。當我思考人與人之間的溝通時，有很多——（of how humans already act in the world when i think about communication between humans so much）
- `18:43` 其中有些是非語言的，像是我手部移動的方式、臉部動作，以及這些你有多大程度能夠（of it is non-verbal just the way i move my hands the facial movements and like how much of that do you）
- `18:50` 預見未來會被人工智慧感知，或被人工智慧讀取（see in the future being sensed by artificial intelligence or read by artificial intelligence）
- `18:56` 也許是透過視覺。這真的重要嗎？因為你現在描述的只有文字，而（maybe through vision is that even important because what you're describing now is text only and）
- `19:01` 對我們這些在網路時代長大的人來說，我們非常習慣透過文字溝通，而且你知道，還會加入（for those of us who grew up online we're very used to communicating over text and you know adding）
- `19:06` 細微之處來傳達我們真正的意思、語氣。可是，讓 AI 能夠讀取我們的臉部表情、手勢等等，這仍然重要嗎？（subtleties to that text to convey what we really mean tone um but like how is it still important to have）
- `19:15` 我想是的。當、當、當（ai be able to read our facial expressions our hand gestures and so on i think so um so when when when）
- `19:22` 我想像我們正在打造的未來，會非常融入環境，也非常自然。如果（i think about the future of what we're building it's it's very ambient it's very natural um if）
- `19:30` 你知道，明天，或是之後某個時候，我去辦公室，在（you know tomorrow uh or like you know later i go i go to my office and i write something on the）
- `19:36` 白板上寫了些東西，突然有個想法，它應該也能夠出現在那裡，而且還能（whiteboard and i have an idea it's like it should be capable of you know being there as well and like）
- `19:41` 理解；或者我可能會跟它說：「嘿，你覺得這件事怎麼樣？」（you know understanding or you know maybe i tell it like you know hey it's just like you know what about）
- `19:44` 然後我們就能直接透過語音進行自然的對話。因為自從（this thing and you know and then we just have a natural conversation just over voice like since）
- `19:49` 我們推出新版 ChatGPT 語音功能後，它真的大受歡迎，所以使用量（we we shipped the new chat gbt voice like the it's it's really taken off so it's like the amount of）
- `19:55` 只透過語音與 ChatGPT 互動的使用者（users that interact with chat gbt just through voice）
- `19:56` 只透過語音與 ChatGPT 互動的使用者，現在成長得非常快，而這個、這個我想其中的（interact with lgbt just through voice is growing very fast right now and uh this is this is i think）
- `20:02` 啟示就是，每當你開始傾向於採用更自然的方式（the lesson is like every time you sort of like lean into something that is more natural）
- `20:07` 人類就是會選擇阻力最小的路徑，你知道，就像你說的，這就像是（like humans just choose the the path of least resistance you know as you said it's like）
- `20:12` 在一個小方框裡打字，你知道，對我們其中一些人來說，這可能很自然，但（typing on a little box like you know it's just like it's natural maybe for some of us but not）
- `20:17` 對每個人來說都不是，而且當你找到某種只是稍微（for for everyone and it's like definitely when you get something that is just like a little bit）
- `20:21` 更容易一點、更好一點的方式，你知道，人們通常就會直接去使用它之類的。對，好（easier a little bit better it's like you know you tend to just go and use that and stuff yeah okay）
- `20:25` 首先恭喜你，我看到你今天早上發布消息，Codex 已經達到兩千萬（i first of all congratulations i saw that you posted this morning codex reached 20 million）
- `20:31` 名使用者。我看過那張圖表，你知道，有一段時間它是這樣，結果突然間變成（users i've seen the graph and and you know for a while it was like this and all of a sudden it's）
- `20:38` 垂直上升。所以恭喜你。我想聊一下和 Anthropic 的競爭（vertical so congratulations i want to talk a little bit about that competition with anthropic）
- `20:43` 因為當然，你知道，很多人認為 OpenAI 和 Anthropic 是（because of course you know a lot of people think open ai anthropic these are the two major competitors）
- `20:48` 目前業界的兩大主要競爭者。曾經有一段時間，Anthropic 幾乎吸走了所有（in the industry right now there was a period of time in which anthropic was kind of sucking all）
- `20:55` 注意力，對吧？他們真的佔據了主導地位，然後突然間有些事情（the oxygen out of the room right they were really dominating and then all of a sudden something）
- `21:00` 改變了。所以首先，你怎麼看現在的市場？對，目前我們真正專注的是（changed uh so first of all what's your read on the market today yeah really right now we're focused on）
- `21:08` 打造能力最強的模型，打造非常、非常高效率的模型，然後投入大量心力（building the most capable models building models that are highly highly efficient and then taking a lot）
- `21:14` 為所有人打造產品。而且我認為，OpenAI 有件事做得（of pride in building products for everyone um and i this is something that i think open ai does）
- `21:21` 非常好，就是關心這個世界，也關心我們要如何把這項非常、非常（uh really well is caring about the world and caring how about you know how we are taking this very very）
- `21:29` 強大的技術交到盡可能多的人手中。這也是你（powerful technology and like putting in the hands of as many people as possible and this is what you）
- `21:34` 知道，我們把 Codex 和 ChatGPT 整合時所做的事。那是一種想法：我們擁有（know we did as well like with merging codex and chat to bt it was like this desire of like we have this）
- `21:39` 這項技術，我們可以讓它更安全，也可以讓它對所有人來說更容易使用（we have this technology we we we can make it safer we can make it easier to to use uh for everyone）
- `21:47` 不管你是產品經理、設計師、業務、行銷、公關，還是其他職位（whether you know you're like a product manager a designer in sales marketing comms all of that）
- `21:54` 你知道，你都應該能夠使用這些功能，然後只是非常、非常快速地，你知道（like you know you should be able to use all of it and then uh just very very quickly you know）
- `22:00` 透過 ChatGPT 發布給大眾，而我們已經有大量使用者，所以這也一直（distributed through chat gbt like where we have a ton of users already and so that's been that's been）
- `22:06` 真正推動著你剛才提到的這波成長，而我不太會一直關注（really driving you know this growth as well that you mentioned and i don't tend to look at the）
- `22:12` 競爭對手。我真的比較關注，我們能夠獨特地把哪些事情做好，以及我們的（competition that much like i really look at you know what can we do uniquely well and what are our）
- `22:17` 價值，以及你知道的，我們要如何最大限度地加速朝那個方向前進。好，我可能想再（values and like you know how do we maximally accelerate towards that okay um i want to maybe）
- `22:22` 稍微深入探討一下，因為我知道你其實沒有那麼常思考 Anthropic（just dig a tiny bit more into that because i i know you're not thinking about anthropic all that much）
- `22:28` 但很多其他人會，他們會想，好吧，我到底相信哪個產品？（but a lot of other people do and they're they're thinking about okay which product do i believe in）
- `22:33` 我想把兩千兩百美元花在哪個產品上？當你觀察市場定位、品牌形象，以及 OpenAI 的語氣（which product do i want to give my twenty two hundred dollars to um when you look at the market）
- `22:39` 還有它與開發者及更廣泛受眾互動的方式時，（position and the branding and the tone from open ai and just the way that it interacts）
- `22:45` 你覺得這和 Anthropic 的做法相比起來如何？（with developers with the broader audience how do you see that comparing to the way that anthropic does）
- `22:50` 對，我想也許再說一次，我很在意的是為全世界建立社群，（yeah i think maybe again like what i care a lot about is like the community building for the world）
- `22:57` 也就是讓每個人都能一起參與。我想，你可以從我們的做法感受到這一點，我們非常（like bringing everyone along um i think you know you can feel that in the way that we we are super）
- `23:04` 透明，像是我們會從社群取得很多想法。老實說，這也真的非常有趣，（transparent about things like we take a lot of ideas from the community it's just like it's also）
- `23:09` 因為我們自己也從中獲得很多能量。然後，（so much fun to be honest um you know because we get so much energy from it as well um and then）
- `23:15` 我們正在打造的這項技術，不是只為了我們自己。我們不是（this technology that we're building we're not building it just for ourselves like we're not）
- `23:19` 只想打造它來加速 OpenAI 而已，這項使命非常重要，（just building it to accelerate um just to open ai it's like it's super important the mission is super）
- `23:25` 所以你知道，這也是我們獲得能量的地方。這一切對我來說（important and therefore it's like you know this is where we also get our energy from um and so it just）
- `23:31` 感覺非常踏實，也很有趣，然後好的事情就會因此發生。（feels to me it feels like very grounded uh it feels fun and then good things happen as a result）
- `23:36` 好，那我們來談談其中一些好事。我想先談一下重置，（of that well let's talk about some of those good things i want to talk about the resets for a second）
- `23:40` Tibo，這算是大家都在關注的事，我知道大家幾乎會追蹤你每一則推文，（tibo that's kind of like i know it's like what everybody is you know kind of following your every tweet）
- `23:46` 就是因為這個。具體來說，再次回頭看 Codex 的成長曲線，這可能是個蠢問題，（because of this um specifically like again looking at that growth curve of codex maybe this is a silly）
- `23:54` 這些重置有多少是對行銷和成長的助益？還是說，（question how much do those resets how much of it is a boon towards marketing and growth or is it）
- `24:00` 它就只是為開發者社群累積善意？我想這可能有點反直覺，但（was it just like goodwill for the developer community i think maybe it's counterintuitive but）
- `24:07` OpenAI 是個非常特別的地方，你可以在這裡直接做事。（open ai is a very it's a place where you can just do things um and so it just felt right initially）
- `24:15` 所以一開始，當我們反覆迭代並弄壞東西時，補償使用者就覺得是對的，或者（to compensate when we were iterating and breaking things or you know maybe we had misconfigured）
- `24:22` 也許是我們把某些東西設定錯了，結果沒有達到我們想要的水準。所以我們會說，嘿，謝謝你來試用（something and it wasn't quite as good as we wanted and so it's like hey you know thank you for trying）
- `24:27` 這項產品。我們知道，我們真的很努力在打造它，現在還處於早期階段。你（this product like we know like we're trying very hard to build it it's like it's early days um you）
- `24:32` 知道，這裡有一些額外用量，因為我們剛好把它弄壞了大概三十分鐘，（know here's some extra usage because you know we happen to break it you know for like 30 minutes and）
- `24:38` 而且我們知道這對你很重要，你也依賴它。謝謝你成為（you know we understand this is like really important and rely on it and you know thank you for being a）
- `24:42` 使用者。所以一開始就是這樣，而我現在仍然是這樣處理：如果我們把它弄壞了，（user and so this is how it started um and you know this is how i still treat it it's like if we break）
- `24:49` 或者使用體驗不理想，而我們又完全不清楚原因，（it um or if the experience is suboptimal and we don't fully understand why it's like you know we will）
- `24:55` 我們就會補償，重置用量上限。然後這件事就變得相當受矚目，（we will compensate for that we will reset um the usage limits and then you know it turned into like）
- `25:00` 現在顯然已經有一個完整的重置按鈕了，不過其實沒有（quite the thing obviously there's a whole reset button now and like but there isn't really a whole）
- `25:05` 太多審核機制。這不是和行銷部門（a lot of scrutiny behind it it's like it's not done in partnership with like marketing）
- `25:09` 或財務部門合作完成的，我想什麼時候按就可以按，（or or finance it's just like i can press the button whenever i want um whenever it feels right）
- `25:15` 只要我覺得時機對了就行。我們有這些原則：我們正在努力打造很棒的東西，當它沒有達到水準時，（um and we have these principles that you know we're trying to build something amazing and when it is）
- `25:21` 我們就會補償。對，我還是覺得其中一部分確實（not it's like you know we will make up for it yeah i i still think there's a piece of it that）
- `25:26` 在社群中建立了很多善意，也可能至少在一小部分程度上促成了成長。（really has built so much goodwill in the community and maybe has contributed to the growth at least in）
- `25:32` 我認為在乎使用者，整個文化確實能發揮很大作用。所以，（a small part i think caring for your users it's all the culture does does a lot right so um i think you）
- `25:39` 你可以口頭上說你在乎，或者你也可以真的在乎。（can pay lip service and say that you care or you know it can be like you know we actually care and）
- `25:43` 如果我們把它弄壞了，就會說，嘿，真的很抱歉，這裡是（like you know if we break it like you know hey really sorry about it you know it's like here's）
- `25:48` 我們彌補的方式。這讓我想起 Amazon 的退貨政策，（here's like how we make up for it it kind of reminds me of amazon's return policy it's like）
- `25:52` 不管怎麼說，只要你不滿意，就儘管把它退回去。而你們其實正在建立（if you're not happy in any sense go ahead and send it back and and you're kind of building that）
- `25:57` 同樣的文化、同樣對 OpenAI 的觀感：嘿，如果我們犯了錯，儘管再使用那些（same culture that same perception of open ai it's like hey if we make a mistake go ahead use those）
- `26:03` token，或者，來，這裡有一批新的 token 給你。對，我真的很欣賞這一點。（tokens again or or you know how here's a here's a fresh batch of tokens for you i yeah i really）
- `26:08` 然後也有一些美好的時刻，是我們只是想慶祝、紀念某個時刻，（appreciate it so and then there is also you know good moments where we just want to celebrate）
- `26:12` 而且實在沒有其他更有意義的東西可以送。（and mark a moment and it's just there isn't really something that we can give that is more meaningful）
- `26:16` 有時候，你知道，我們總是會推出新功能，也會盡可能廣泛地推出，（um you know at times like you know it's like we always ship new features we will ship them as）
- `26:21` 但要和整個社群分享某件事，就像是，嘿，去探索這個新東西吧。（broadly as we can um but something to share with the entire community it's like you know hey go）
- `26:26` 你還沒用過 Ultra，對吧？這裡有一些額外用量，（explore this new thing like you know it's just like you haven't used ultra yet you know here's）
- `26:29` 好好試試看。我聽說現在真的有一個實體按鈕了。（some extra usage like you know good try it and and i heard there's an actual physical button now）
- `26:35` 對，有。好吧，你之後得讓我看看。我會拿給你看，真的非常非常（yes there is yeah okay you'll have to show me that after i will show it to you it's like very very）
- `26:40` 酷。不過，進行這麼多次重置，只有在你做過（cool but with all of these resets like you can really only do that if you've done）
- `26:45` 大規模的運算容量規劃時才有可能做到。你必須有足夠的運算資源，才能提供所有這些（significant compute capacity planning like you have to have enough compute to give all of these）
- `26:50` 重新開始，我想先稍微談談自我改進，因為，說到（resets and i i want to start to talk a little bit about self-improvement because um like speaking of）
- `26:56` 算力，幾週前，我想應該是幾週前，你發布了一篇文章，裡面（capacity a few weeks ago i think it was a few weeks ago there was this uh article you put out and it）
- `27:03` 提到 OpenAI 優化了 Luna 的效率，把 Luna 的價格降了 80%，另外 Tara 也降價了（stated soul had optimized luna efficiency you dropped the price of luna by 80 there was also a price drop）
- `27:11` 你們從 Luna 擠出了多少效率，又有多少是因為我們真的做了非常好的算力規劃，然後可以直接把（for tara as well how much of a an uh efficiency game were you able to eke out of luna versus how）
- `27:18` 價格降下來，因為我們的利潤還是很好，而且我們也希望大家使用它？其中有多少來自（much of it is like we just did really great compute capacity planning and and we can just drop the）
- `27:25` 演算法上的提升，又有多少來自策略規劃？我們很早就開始規劃算力了。你知道，我想如果回頭看兩年前，我覺得 OpenAI 當時曾被質疑，為什麼要在算力上投入（price like our margins are great and we can still we want people to use it so like how much of it were）
- `27:30` 這麼多資源。這真的是一個非常明智的投資。是的，現在我們很樂意讓非常大一部分的電腦資源用於研究，投資我們的未來，持續打造（algorithmic gains versus um strategic planning we planned uh compute like way ahead you know i think）
- `27:37` 更好的模型，同時也提升現有模型的效率。（if you look back two years i think open ai was um kind of questioned for why you know there was like）
- `27:45` 而現在正在發生的奇妙事情是，當我們推進現有最先進模型的能力前沿時，就能利用這些模型，非常（so much investment in compute uh one of those crazy good bets yes uh and then now we're very happy to）
- `27:52` 快速地找出該如何提供服務、重構，或重新設計我們的技術堆疊，以獲得非常（have it like a very large fraction of the computers used for research where we invest in our future and）
- `27:57` 顯著的效率或效能提升。所以我們不只是改善了這項功能。這件事我們也會發表相關內容。我們不只是提升了成本效率，也提升了速度（you know ever ever better models and then also like the efficiency of the models that we have）
- `28:03` 效率。你知道，除了超高速的部分之外，整體速度也隨著時間顯著提升了。（and then the amazing thing that's happening is like when we push the frontier of capability for）
- `28:10` 如果你把它畫成圖，就會看到現在獲得的速度大約比三個月前快了 60%。（like the most advanced models that we have then we can use these models in order to figure out very）
- `28:17` 這只是因為我們持續優化技術堆疊的每一個部分，確保我們針對目前的工作負載，以最佳方式進行設計和工程實作。（very quickly how to serve or how to restructure uh or re-engineer our stack in order to gain very）
- `28:25` 而且，我們最強大的模型，正是讓我們能以非常小的團隊做到這些事情的模型。（significant efficiency or performance gains so we haven't just improved this is something that we will）
- `28:30` 所以，每當我們取得非常顯著的效率提升和成本效率改善時，我們的承諾就是持續讓一切維持在效能與成本的前沿，同時也不只是（publish on as well we haven't just improved the cost efficiency but we have also improved the speed）
- `28:35` 效率方面，你知道，除了超高速的事情之外，其他都已經隨著時間顯著加快了。（efficiency you know outside of ultra fast things have gotten significantly faster over time they have）
- `28:40` 如果你把它畫成圖，就會看到現在你所獲得的速度大概是（like if you plot it it's like you know the amount of um just the amount of speed that you get now is）
- `28:45` 比三個月前快了大約百分之六十。（like you know roughly 60 percent faster than you know what it used to be like three months ago）
- `28:50` 而我們現在就是在處理整個技術堆疊的每個部分，確保（and this is just like we're we're just going after every part of the stack and just really making sure）
- `28:57` 我們針對目前的工作負載，以最佳方式進行設計與工程實作。所以，你知道，（that we design it and engineering it optimally for the kind of workloads that we have and so you know）
- `29:03` 我們擁有的最強大模型，就是那些真正讓這件事變得（and the most powerful models that that we have are the ones like you know just really that make it）
- `29:09` 可行的模型，讓我們只靠一個非常小的團隊也能做到。所以，大多數時候，當我們（capable for us to do it you know with a very small team and so the majority of like what when whenever we）
- `29:15` 取得非常顯著的效率提升和成本效益時，我們的承諾就是（come up with like very significant efficiency gains and cost efficiency like our commitment is to just）
- `29:21` 持續維持在效能與成本的最前沿，同時也不（really to keep things at the frontier of performance cost um and to just also like you know just not）
- `29:30` 就先把它放在口袋裡，你知道，如果有有趣的成果，就把它做成我們知道可以分享給（just pocket you know that uh if interesting gain and just make it something that we know we share with）
- `29:35` 客戶、分享給使用者的東西，這就是我們在 Luna 上做的事。那你覺得（our customers who share with our users and that's what we did with luna how do you what do the）
- `29:40` 內部討論會是什麼樣子？當你們試著決定，要把多少運算資源分配給研究新（discussions look like internally where you're trying to decide compute allocation towards researching new）
- `29:46` 模型、改善既有模型的效率，以及推論時的效率時，內部會出現什麼樣的拉鋸？（models efficiency gains on existing models inference like what does that tension look like internally）
- `29:52` 嗯。（um）
- `29:54` 我們通常會從第一性原理的角度來看事情，而且我們會為研究分配一部分（the we we usually look at things um from from first principles and we have like an allocation）
- `30:02` 資源，也會為產品分配一部分資源，接著在產品內部做出不同（for research we have an allocation for um for product and then within product we make different）
- `30:08` 種類的取捨。不過這次幾乎算不上取捨，因為效率（kinds of trade-offs but this one was uh almost not even a trade-off because uh the the efficiency）
- `30:15` 提升本來就存在，所以我們基本上能夠在相同的運算資源範圍內，（gains were there um so you know we were pretty much like able to use like the same compute envelope）
- `30:19` 提供非常、非常顯著的吞吐量提升。對，所以當（in order to you know serve this very very uh very significant increase in throughput yeah so when）
- `30:26` 我的意思是，幾個月前我看到你們那篇部落格文章，在降價公告之前，你們在文章中（i mean when i saw the blog post a few months ago prior to the price drop blog post where you guys）
- `30:33` 提到用一個模型訓練下一個模型，或協助最佳化下一個模型，（were talking about one model training the next model or helping kind of optimize the next model）
- `30:37` 然後你看到 Soul 透過觀察 Luna 的運作方式，所達成的這些效率提升。（um then you see these efficiency gains that were achieved by soul looking at how luna was running）
- `30:44` 呃，我是說，在我看來，遞迴自我改進才剛開始沒多久。（uh i you know it seems to me like recursive self-improvement in the very early innings）
- `30:49` 你覺得呢？現在發生的事情是遞迴自我改進嗎？嗯，我想遞迴自我改進（what what are your thoughts there is that what is happening um yeah i think recursive self-improvement）
- `30:55` 顯然是現在非常熱門的議題，而且我想最常被套用在研究上，（is uh it's obviously a huge topic right now and it's most often i think applied to to research）
- `31:02` 嗯，也就是讓模型開發其他模型；但我們看到非常多成功案例的是，（um and you know models developing other models but what we are seeing a ton of success with is）
- `31:10` 你知道，使用那些模型來開發處於關鍵路徑上的基礎架構，（you know using those models to develop the infrastructure that is on the critical path）
- `31:14` 也就是使用那些模型本身，這同樣也是遞迴自我改進的一種形式。所以這其實是同一個（of using those models you know which is also a form of recursive self-improvement so it's all one）
- `31:18` 大型系統，從前端到堆疊。你知道，那些更底層、更困難的核心，以及我們使用的 CUDA 核心，（big system in front stack you know the the harder the the the kernels uh cuda kernels that we use）
- `31:26` 嗯，開發新產品，以及更有效率地與那些模型互動的新方式，（um developing new products and new ways to interact with those models that are more efficient）
- `31:32` 你提到雲端 Agent；如果我們真的破解了雲端 Agent，情況就像是（you know you talked about cloud agents it's like if we if we really crack uh cloud agents it's like）
- `31:39` 你突然也變得更有效率了。這算不算是一種遞迴自我改進？（suddenly you become much more productive as well it's like is that a form of like recursive self-improvement）
- `31:43` 因為這樣一來，你就更有能力從它們身上取得效用。我也認為，在（because then you know you have a better ability to get the utility from them also i think it is in）
- `31:48` 某種程度上是，但它更偏向基礎架構，接著就是能夠（some sense but it's much more um you know infrastructure and then you know being able to）
- `31:52` 然後把它拿來，再把它指回自己，對，所以當然，我們就是這麼做的。（then take that and then point it back at itself yeah and so of course like we're doing that it's）
- `31:57` 如果我們沒有這麼做，我覺得那會蠻蠢的。你可以稍微談談（like if we were not doing that um i think that would be pretty silly can you talk a little bit about）
- `32:03` 既然我們談到遞迴自我改進，OpenAI 的 Sam Altman 最近談到要暫停目前最前沿的強化學習，我想應該是這樣。你可以稍微談談那件事嗎？當時是怎麼（so as we're on the topic of recursive self-improvement uh open ai sam altman talked about pausing the）
- `32:09` 做出那個決定的？（absolute frontier of rl right now i believe um can you talk a little bit about that like what was）
- `32:13` 我知道我們剛剛簡短談過 Hugging Face 的事件，但究竟是什麼促成了那個決定？那看起來是什麼樣子？那些討論是怎麼進行的？（that decision like and i know we talked about the hugging face incident briefly but like what went）
- `32:20` 對，這是研究領域裡非常重要的一件事，OpenAI 一直都能把資源投入在最重要的地方。（into that decision what does that look like how did those discussions go yeah this is this is something）
- `32:26` 而且隨著我們的模型能力不斷提升，（very much within within research where there is um open ai has always been able to invest）
- `32:34` 很明顯，對齊和安全性方面也變得越來越重要。因此，投入非常大量的資源在這方面，對 OpenAI 來說是非常自然的事，（uh its resources where it matters most um and as we increase the capabilities of our models it is very）
- `32:42` 也是 OpenAI 非常重視、非常投入的事情。所以我們看到，這方面的投資出現了大幅增加。（obvious that you know the alignment and the safety aspect of it is you know ever more important and so）
- `32:48` 而且那次暫停某種程度上是必要的，為了讓團隊和個人能夠真正理解並強化系統的各個部分，（having uh having tremendous um amount of investment there uh is is a very natural thing for open ai and）
- `32:57` 接著才能確保（like something that open ai is very committed to and so we're seeing um a huge surge uh in investment）
- `33:05` 它們能夠……（um on this and also the uh the the pause was sort of like necessary to uh allow like the teams and）
- `33:15` 確保（individuals like just really understand and harden all parts of the system uh to then you know ensure）
- `33:21` 也就是說，我們可以重新開始訓練，讓你擁有完整、完整、完整的控制權，然後這件事——（that you know we could we could restart training uh with you know like full full full command and this）
- `33:27` 我相信 OpenAI 會一直持續這麼做，必要時就會這麼做。我——（is something that you know i believe open ai will always continue to do like when when necessary i've）
- `33:32` 我沒看過我們內部無法非常有效率地做出這類決定。（i've i've not seen i've not seen us internally not able to make such decisions like very efficiently）
- `33:39` 當時是否有設定某個目標，而且很明確地知道，在解除暫停之前必須先達到這個階段？（was there like some set goal in place where it was very clear you needed to reach this point before）
- `33:45` 還是說，是抱持著『看到時就會知道』的想法？對，這件事是由安全團隊負責，嗯，他們——（unpausing or was it hey we'll know it when we see it yeah this this is something that sits um within）
- `33:52` 他們非常重視這件事，這很大程度上是一場辯論，也是一個邊做邊探索的過程。（within the the safety team and uh they they very much this is like very much a a debate and sort of）
- `34:01` 但後來他們確實達成了一套相當明確的原則，嗯，只要達到這些原則，我們就會處在一個不錯的位置。（like a discovery process as you go um but then they did reach uh a fairly clear set of principles）
- `34:08` 我想回到超高速模式。我覺得大家沒有意識到，這種速度能解鎖什麼樣的可能性。所以我們先從使用情境談起：你們內部有哪些使用方式，是在擁有這種每秒 token 數之前做不到的？（uh that you know when reached like you know we would be in a good position i want to go back to）
- `34:13` 我們看到它在利害關係很高的情況下被大量使用。舉例來說，當你們——（ultra fast mode that i think people don't appreciate what that kind of speed unlocks and so let's start）
- `34:21` 當我們遇到服務中斷時，事故指揮官和回應團隊就能使用超高速模式。（with what use cases are you doing are you using internally that were not possible prior to having）
- `34:27` 我們看到它在利害關係很高的情況下被大量使用。舉例來說，當你們——（those kind of tokens per second we see it used a lot when the stakes are high um so for example when you）
- `34:35` 當我們遇到服務中斷時，事故指揮官和回應團隊就能使用超高速模式。（have um when we have an outage um the incident commander and the response team gets access to ultra fast）
- `34:44` 嗯，因為每一秒都很重要，所以在高風險、高 stakes 的情境下，就像是（um because so every every second is you know matters um so high stake um high stakes scenarios like just）
- `34:55` 真的不太會使用 Ultra Fast，而且這算是一件蠻有趣的事，因為有些團隊（really weren't you know using ultra fast also um it's it's kind of like a fun thing where uh teams）
- `35:04` 不是正在處理非常關鍵的事情，就是認為自己正在處理非常關鍵的事情，（which are like either working on something very critical or believe they are working on something）
- `35:09` 也都會要求使用 Ultra Fast。那寵物算在這種情況裡嗎？寵物？對，寵物（very critical will always request ultra fast as well um does pets fall under that uh pets yeah pets）
- `35:17` 寵物還不到極度關鍵，但我很愛、很愛我的寵物。牠總是在我的螢幕上，像是你走（pets is not quite hypercritical but i love i love my pet it's always on my screen uh like when you walk）
- `35:23` 來走去時，會看到大家的寵物出現在螢幕上，而且當他們撥入（around uh you see like you know people's pets on their screen and like also when they they dial in）
- `35:29` 視訊通話時，也總是會看到。這讓我覺得非常愉快，每次看到都會帶給我喜悅，（into the the video call it's just like it always like i think it's very delightful and it brings me joy）
- `35:35` 但寵物現在還不算關鍵。我們確實會維護它，也會好好照顧我們的寵物，（every time i see it but uh pet is not quite critical right now um we do maintain it um and we take good）
- `35:42` 但假設有人正在研究一個他們想到的新點子，而且他們（care of our pets but uh say you know someone is working uh on like a a new idea they have and they're）
- `35:50` 會說：「嘿，我真的覺得這可能會是很特別的東西。」我們（like you know hey it's just like you know i really think this could be like something special and we）
- `35:55` 得試試看，但我們必須在星期一決定，要不要把這個放進 Dev Day，（have to try it but like you know we have to make a decision on monday on like you know whether we include）
- `35:59` 那就會想：「好，當然就使用 Ultra Fast。」（this in dev day or not and it's like okay just like you know of course you know use ultra fast um）
- `36:04` 大家對於自己喜歡像我們之前談到的單執行緒，還是很常多工，會有不同偏好。（people have different kinds of preferences on uh whether whether they like to be you know monothreaded）
- `36:11` 對於很喜歡多工的人來說，你從 Ultra Fast 得到的好處就沒那麼多，對吧？但有些人（as we talked about or multitask a lot for folks who like to multitask a lot you don't benefit as much）
- `36:17` 不喜歡一直切換情境。你在這個光譜上比較接近哪一端？（from ultra fast right but some people which like don't like to change context all the time um）
- `36:23` 我有 ADHD，所以我一直都在切換情境。（where do you fall on that spectrum i i have adhd so i like i context switch like all the time）
- `36:29` 很有趣，因為我也有 ADHD，但其實我不想一直切換情境，那對我來說（it's funny because i also have adhd and i actually don't want to context switch all the time that's）
- `36:35` 真的很難。我想專注在兩到三件事上，所以我才會這麼期待 Ultra Fast。（really hard for me i want to focus on two to three and that's why i was so excited about ultra fast）
- `36:40` 真有意思，你剛好相反。你知道，我很享受切換情境。（that's fascinating that you're the opposite there you know i thrive in context switching）
- `36:44` 好，做很多小決定。不過你知道，有時候我確實只想專注在（okay making lots of little decisions um but you know sometimes i do want to just stay focused on）
- `36:51` 一件事上，這時 Ultra Fast 就非常棒，因為它會讓你一直保持在（like one thing and then ultra fast is just delightful because it just keeps you just right there in the）
- `36:55` 心流裡。Ultra Fast 的特點是，當不需要呼叫太多工具，或是需要大量產生（flow the thing with ultra fast uh that you know we we it works amazingly well when there's not that）
- `37:02` 情境內容時，它的效果非常驚人。比如說，如果你想要製作原型，像是（many tool calls involved or it's like a lot of generation of context so for example if you're）
- `37:08` 一個網站或電子遊戲，而你只需要它撰寫大量（trying to prototype um a website or a video game and you know you just need to it to write like a lot of）
- `37:14` 程式碼，它就會快非常多，對吧？大概快十倍。但如果它是（code um then it will do it so so quickly right you know 10 times more quickly but if it's a）
- `37:21` 很多工具呼叫的額外開銷，其實是在網路的其他地方，或是在 Agent 執行軌跡的其他地方。（lot of tool calls like the overhead is in like somewhere else in the network or you know somewhere）
- `37:25` 所以你只會感覺到大約 3 倍或 4 倍的加速，對，你不會得到完整的 14 倍加速。所以我知道 OpenAI 的員工有無限量的 token，我可以想像，如果我有無限量的 token，我一定會把它調到最大，想著用 5.6，Solar，或是最新的模型。（else in the agent trajectory then you know you only only feel like a 3x or 4x speed up yeah um you'll）
- `37:32` 你不會得到完整的 14 倍加速。所以我知道 OpenAI 的員工有無限量的 token，我可以想像，如果我有無限量的 token，我一定會把它調到最大，想著用 5.6，Solar，或是最新的模型。（not get that for like full 14x so i i know open ai employees get unlimited tokens and i can imagine if）
- `37:39` 我知道 OpenAI 的員工有無限量的 token，我可以想像，如果我有無限量的 token，我一定會把它調到最大，想著用 5.6，Solar，或是最新的模型。（i had unlimited tokens i would always set it to max thinking 5.6 solar whatever the latest model is）
- `37:45` 我也會有類似的想法，我總是會希望使用 ultra fast，就像在不必考慮成本時，（and i would think kind of similarly i would always want ultra fast on it's like when when cost isn't）
- `37:52` 我會想，好，那就把它開到最大。內部也是這樣嗎？我們不會把 ultra fast 提供給每個人，（on my mind i'm like okay max it out is that how it is internally we don't we don't give ultra fast to）
- `37:58` 我們會保留很多容量給外部使用者和客戶。對，嗯，所以 OpenAI（everyone like we reserve a lot of our capacity for external users and customers yeah um so open ai）
- `38:05` 員工具備把所有這些資源都吃下來的能力和容量，對吧？也就是說，像我們所有的（employees have the ability and the capacity to gobble up all of it right so gobble up like all of our）
- `38:12` 生產環境 GPU，所有超高速的 GPU，我們其實會全部用上，但你知道，我們不會（production gpus all of ultra fast is like uh you know we would use all of it but you know we don't）
- `38:19` 完全這麼做。我們會以某種方式限制用量，像是評估對我們來說多少才合理，（like we sort of um we restrict it in a way uh where you know like we we look at you know how much）
- `38:26` 這樣我們就能使用這些資源，也能充分了解產品，進而（is reasonable for us yeah so that you know we use it so that we understand the product as well so that）
- `38:31` 持續改進產品，從遞迴自我改進中受益。不過，絕大多數資源都會（we keep improving it so that we benefit from you know recursive self-improvement but um the the vast）
- `38:37` 保留給客戶。好的，沒錯，謝謝。那麼，哪些情況對延遲比較敏感？（majority is like reserves for customers okay yeah that's good thanks um uh what latency sensitive）
- `38:44` 在 OpenAI 以外，還有哪些應用情境是你最期待的？這種速度會解鎖哪些可能性？（use cases outside of open ai are you most excited about that gets unlocked by that kind of that kind）
- `38:50` 這種速度真的很有意思。我現在最期待的一件事，總的來說就是（of speed it's interesting it's just like really one thing that i'm very excited about in general is）
- `38:57` 非文字互動。比如說，你能不能在共享畫布上操作？你能不能（um non-text interactions so um can you can you like sort of operate on a shared canvas can you）
- `39:06` 建立東西？你能不能做到、能不能產生一些想法，或是產生不同的（create things can you do can you generate you know ideas and different uh can you generate different）
- `39:12` 圖片，然後選出其中一張，就像……你知道，選擇你的冒險一樣，接著你（images and then select one and like so like you know choose your adventure uh and and and then you）
- `39:17` 很快就能得到一個原型的雛形，然後你可以即時引導它（know have a very quick mock-up of a prototype that then you can steer um you know like in real time）
- `39:23` 透過語音或文字來調整，接著你就能直接在眼前看到成果。這就像是一個（matter through voice or through text and then you sort of like just see it right there um it's like）
- `39:28` 非常有創意的過程，我覺得這種速度讓它成為可能。你知道，身為工程師，（this very creative process which i think these speeds allow um where you know like as as an engineer）
- `39:36` 有時候你會先退一步，心想：「噢，我需要設計整個系統，（sometimes you know you're just like sort of like you you sit back and you're like oh i need to design）
- `39:39` 我需要思考取捨和需求。」但也許你（this whole system i need to think about it the trade-off the requirements but like you know maybe）
- `39:44` 只要一分鐘就能把它做出來，看看實際運作的情況。對，然後就能（you know you can just create it in one minute and see like how it actually does yeah um and then sort）
- `39:50` 更投入其中、更順著思路進行，也能更了解事情。（like be more like in the flow and like you know into it things better and）
- `39:52` 低延遲，而且你知道，Intuit 的表現更好。（low and like, you know, Intuit thinks better.）
- `39:54` 而且我覺得，這些速度讓它能夠發揮作用。（And I think these speeds allow it out.）
- `39:55` 對。（Yeah.）
- `39:55` 所以（And so）
- `39:56` 我猜超高速的價格會比一般的速度高上不少。（I'm assuming the ultra fast price is going to be significantly higher than than kind of normal）
- `40:01` 速度。（speeds.）
- `40:01` 你覺得超高速會成為標準，還是它們永遠都會維持較高的價格？（Do you think ultra fast speeds are going to become the standard or are they always going）
- `40:07` 收取溢價？（to have a premium price point?）
- `40:09` 嗯，這很有意思。（Um, that's interesting.）
- `40:11` 所以我覺得，這就跟（So I think the same way as）
- `40:13` 科技通常的發展方式一樣，（technology usually goes,）
- `40:14` 我覺得隨著時間過去，它會變得更普及、更普及，（I think it will become like more broadly and broader and broader）
- `40:18` 也更容易取得。（accessibility over time.）
- `40:20` Agent 完成工作的速度，你知道，（Um, the speeds at which like agents get things done, like, you know,）
- `40:26` 會持續提升。（will continue to improve.）
- `40:27` 我們看到每個月都有大幅的進步。（Like we're seeing massive improvements like month after month.）
- `40:30` 嗯，這不只是推論速度。（Um, this is not just the inference speed.）
- `40:32` 這也包括模型的 token 效率，也就是它使用 token 的效率。（This is also the, just how token efficient the models are.）
- `40:36` 比如說，Sol 的 token 效率就明顯比 Terra 高。（Like Sol is like significantly more token efficient than Terra.）
- `40:39` 下一個模型的效率會明顯更高，（Next model will be significantly more）
- `40:41` 嗯，token 效率也會比 Sol 高，這應該是你預期的。（efficient, uh, token efficient than, than, than Sol, as you might expect.）
- `40:45` 而且我們一直在持續推進（And we always pushing on）
- `40:46` 這件事。（that.）
- `40:47` 所以隨著時間過去，所有事情都會變快，推論、硬體，你知道，（And so things just get faster over time, inference, hardware, like, you know,）
- `40:51` 一切都是如此。（everything,）
- `40:52` 你知道，就像我們持續在那方面創新，速度也會越來越快。（you know, just like we continue to innovate there and it gets faster.）
- `40:55` 所以我確實覺得，嗯，你知道，（So I do think in, you know,）
- `40:57` 也許一、兩年後，這些速度會變成，你知道，（maybe a year or two, these speeds will become, you know,）
- `41:00` 也許就算不是預設，至少也會非常接近（maybe if not the default, like very close to）
- `41:02` 預設速度。（the default.）
- `41:03` 但我也確實覺得，你知道，你永遠會有高一個等級的選項，（But then I do also think, you know, you will always have like the one tier up,）
- `41:08` 嗯，就是，（um, where,）
- `41:09` 你知道，你永遠可以使用更多硬體。（you know, you can always use more hardware.）
- `41:11` 你永遠可以做出不同的取捨，比如說（You can always do different trade-offs that are like）
- `41:13` 成本更高。但這就只是讓你多得到一些額外的東西。（more costly. Um, but that just kind of gives you something, something extra.）
- `41:18` 所以 Thibaut，我通常喜歡用來收尾的最後一個問題，嗯，是，（So Thibaut, the, the last question I usually like to end on, uh, is,）
- `41:22` 是針對更廣大的觀眾。（is for a broader audience.）
- `41:24` 外面有很多人對 AI 感到相當不安，不管是因為，（There are a lot of people out there who are quite nervous about AI, whether it's,）
- `41:29` 嗯，工作自動化，（uh, job automation,）
- `41:31` 環境影響，嗯，或者只是對這件正在發生的事情，（environmental impact, um, or just kind of this, this thing that's happening.）
- `41:35` 它，而且感覺（It's, and it feels）
- `41:36` 相當陌生。（quite foreign.）
- `41:37` 你會給更廣大觀眾什麼樣的鼓勵？（What words of encouragement would you give to the broader audience?）
- `41:42` 嗯。（Yeah.）
- `41:42` 所以我們真的打造了面向全世界的 ChatGPT，而且我們非常、（So we, we, we really built for the world with, with ChatGPT and we are very,）
- `41:52` 非常重視，（very much,）
- `41:53` 嗯，投入資源提升它的效率。（um, investing in how efficient it is.）
- `41:56` 而且，你知道，這和我們提供的，像是，（And, you know, this is directly aligned with like,）
- `41:59` 你知道，廣泛存取和廣泛實用性，是直接一致的。（you know, broad access and broad utility that we provide.）
- `42:02` 所以它越便宜，你知道，（So the cheaper it is, you know,）
- `42:04` 要提供服務，比如說，你知道，你能用它做的事情就越多，嗯，（to serve, like, you know, the more, the more you can do with it, um,）
- `42:07` 你從中獲得的也越多（the more you get out of it）
- `42:09` 在你的日常生活中。（in your daily life.）
- `42:10` 而且它已經變得非常、非常有效率。（And it has gotten very, very efficient.）
- `42:12` 比如說，如果你看看，嗯，你知道，（Like if you look at, uh, you know,）
- `42:15` 以 Luna 為例，它是個小得多的模型。（Luna, for example, like it's, it's a much smaller model.）
- `42:18` 它非常、非常有效率，但是（It is, uh, it is incredibly efficient, but）
- `42:20` 如果你回頭看六、六個月前，它當時會處於最前沿。（like if you rewind six, six months ago, it would have sat at the frontier.）
- `42:24` 對。（Yeah.）
- `42:25` 嗯，而你看看（Um, and you look）
- `42:26` Luna 的成本，對吧？（at the cost of Luna, right?）
- `42:27` 就像，你知道，這真的便宜得不可思議。（It's like, you know, it's like, it's, it's, it's crazy cheap.）
- `42:30` 它，（It's,）
- `42:30` 它，嗯，真的非常驚人，對吧？（it's, um, it's phenomenal, right?）
- `42:31` 就是一種令人難以置信的感覺。（It's like a kind of incredible.）
- `42:33` 你們現在還免費提供。（You're giving it away for free now.）
- `42:34` 對，沒錯，它現在就像是在，嗯，嗯，這個免費模式裡，（There, yeah, there, it's just like on, on, on, on, um, um, in this free mode,）
- `42:40` 對吧？（right?）
- `42:40` 嗯，（Um,）
- `42:41` 這就會讓人覺得，哇，你知道，（which is like, wow, you know,）
- `42:42` 這種取得驚人智慧的管道將會變得（it's like this access to incredible intelligence will become）
- `42:46` 無所不在。（like ubiquitous.）
- `42:47` 嗯，而這只有在你不斷推動效率提升時才有可能，我們就這樣，持續提升效率，（Um, and it's only possible when you push, you know, we push the efficiency,）
- `42:52` 你知道，一個月接著一個月、一年接著一年。（like, you know, like month after month, after month, year after year.）
- `42:55` 所以我覺得，你知道，（And so I think, you know,）
- `42:56` 不管以前是什麼樣子，你知道，現在前沿技術就像是，嗯，你知道，（whatever it was like, you know, is a frontier now is like, you know,）
- `42:59` 會變得非常，（will become like way,）
- `43:00` 非常便宜，六個月後就能執行。所以這就是，這就是我的，（way cheaper to run in six months. And so this is, this is like my,）
- `43:03` 呃，大概，你知道，這就是我回答這個問題的方式，就是，呃，（uh, sort of, you know, this is how I would answer this question is just, uh,）
- `43:09` 科技有一種方式，（technology has a way）
- `43:10` 會隨著時間變得非常、非常有效率。（to become like, you know, very, very efficient over time.）
- `43:13` 嗯，而我們非常重視，（Um, and we're very focused on like,）
- `43:15` 你知道，讓所有人都能廣泛使用，並且我們在最佳化，你知道，（you know, very broad access and we're optimizing for, you know,）
- `43:17` 你能從中獲得的效用，（the utility that you get out of）
- `43:18` 也就是直接獲得的效用。（it directly.）
- `43:19` 那對於那些甚至對第一次嘗試人工智慧都感到卻步的人呢？（And how about for people who are apprehensive to even try AI for the first time,）
- `43:23` 你會怎麼告訴他們？還有，還有，（like what, what are you telling them and, and how,）
- `43:26` 你要如何描繪、如何描繪出一個，（how can you paint and picture a vision of the）
- `43:28` 人工智慧正在幫助世界的未來藍圖？（future in which AI is, is helping the world?）
- `43:32` 好的。（Yes.）
- `43:32` 嗯，我想，其實你不用看，（Um, I think it's, you don't have to look）
- `43:35` 太遠，你看，ChatGPT 以非常個人、非常深入的方式幫助人們。（very far, like ChatGPT helps, um, people in very personal and deep ways.）
- `43:41` 像是很多，嗯，很多，（Like a lot of, um, a lot）
- `43:42` 使用者會用它來協助寫作，但也會用來，你知道，（of our users use it for help in writing, but also like, you know,）
- `43:46` 尋求個人建議，或是，你知道，（for personal advice or, you know,）
- `43:48` 醫療建議。像是我們推出了，嗯，健康和財務功能，而你知道，我，我，（medical advice, like we, we launched, um, health and finance and, you know, I, I,）
- `43:53` 我自己也非常，（I use them super）
- `43:55` 常用它們，而且，呃，我，我覺得我得到了很多，呃，（regularly and, uh, I, I feel like I get like a lot of, uh,）
- `43:58` 原本，你知道，（support that I otherwise like, you know,）
- `44:00` 我很難獲得的支援。（it would be hard for me to get.）
- `44:02` 例如，這讓我在去看，（And it allows me, for example, to be more informed when I go see my）
- `44:05` 醫生時，能掌握更多資訊。（doctor.）
- `44:05` 所以你不需要走得很遠，嗯，你知道，（And so you don't, you don't need to go very far, um, you know,）
- `44:08` 就能了解它能提供的效用，（to kind of see the utility）
- `44:10` 我想，只要和其他人聊聊，然後，（that it can provide just, I think, you know, talking to others and then,）
- `44:13` 你知道，從，（you know, getting）
- `44:14` 別人使用它的方式，還有，你知道，（inspired by, you know,）
- `44:15` 他們如何從中受益得到啟發，就是一個很棒的方法，可以，你知道，（how others use it and benefit from it is like a great way to, you know,）
- `44:19` 也許，嗯，開始思考你可以如何從中受益。（maybe, um, like start considering how you could benefit from it.）
- `44:24` 好的，Tibo，非常謝謝你。（Well, Tibo, thank you so much.）
- `44:26` 很感謝你撥冗接受訪問。（Appreciate your time.）

## 標籤

訪談, Codex, ChatGPT, 產業趨勢, DeepMind, OpenAI, Anthropic