拉片視頻鏈接傳送門(mén)：

講者是來(lái)自O(shè)penAI（之前在Anthropic）的研究員Karina Nguyen，她參與Claude1，ChatGPT Canvas，Tasks方面的工作。

下面是完整的演講內(nèi)容，“親愛(ài)的數(shù)據(jù)”整理成逐句中英文對(duì)譯：

I worked at antarik for about two years, working on cloud.
我在 antarik 工作了大約兩年，專注于云相關(guān)工作。

Today, I would love to chat more about the scaling paradigms that have happened in the past two to four years in AI research, and how these paradigms unlocked new frontier product research.
今天，我想聊聊過(guò)去兩到四年在人工智能研究中出現(xiàn)的擴(kuò)展范式（Scaling Paradigms），以及這些范式如何開(kāi)啟了全新的前沿產(chǎn)品研究。

I’m also going to share some of the lessons learned by developing Claude and ChatGPT products, some design challenges and lessons, and how I think about the future of agents as they evolve from collaborators to co-innovators.
我也會(huì)分享開(kāi)發(fā) Claude 和 ChatGPT 產(chǎn)品過(guò)程中獲得的一些經(jīng)驗(yàn)教訓(xùn)、設(shè)計(jì)挑戰(zhàn)，以及我如何看待智能體（Agents）從“合作者”演變?yōu)椤肮餐瑒?chuàng)新者”（Co-innovators）的未來(lái)。

In the future, I would also love to invite you to engage in the conversation, and I’d be more than happy to answer some questions at the end.
之后也希望你們加入討論，我很樂(lè)意在最后回答你們的問(wèn)題。

I think there are two scaling paradigms that happened in AI research over the past few years.
我認(rèn)為過(guò)去幾年在 AI 研究中出現(xiàn)了兩種擴(kuò)展范式。

The first paradigm is next-token prediction, also called pre-training.
第一種范式是下一個(gè)詞（token）的預(yù)測(cè)，也稱為“預(yù)訓(xùn)練”（Pre-training）。

What’s amazing about next-token prediction is that it's essentially a world-building machine.
下一個(gè)詞預(yù)測(cè)之所以令人驚嘆，是因?yàn)樗举|(zhì)上是一個(gè)“世界構(gòu)建機(jī)”（World-building Machine）。

The model learns to understand the world by predicting the next word, fundamentally because certain sequences are caused by initial actions which are irreversible, so the model learns some of the physics of the world.
模型通過(guò)預(yù)測(cè)下一個(gè)詞來(lái)理解世界，本質(zhì)上是因?yàn)槟承┬蛄校⊿equence）是由初始動(dòng)作（Initial Actions）引起的，這種因果關(guān)系是不可逆的，所以模型能夠?qū)W到世界的一些物理規(guī)律。

Tokens can be anything—strings, words, pixels—so the model must understand how the world works to predict what's next.
詞（Token）可以是任意的東西，比如字符串（Strings）、單詞（Words）、像素（Pixels）等，因此模型必須理解世界的運(yùn)行方式才能預(yù)測(cè)接下來(lái)會(huì)發(fā)生什么。

Next-token prediction is essentially massive multitask learning.
下一個(gè)詞的預(yù)測(cè)本質(zhì)上是大規(guī)模的多任務(wù)學(xué)習(xí)（Massive Multitask Learning）。

During pre-training, some tasks are easy, such as translation, while others, like physics, problem-solving generation, logical expressions, and spatial reasoning, are very hard.
在預(yù)訓(xùn)練期間，有些任務(wù)很容易，例如翻譯；而另一些任務(wù)，如物理知識(shí)、問(wèn)題求解生成、邏輯表達(dá)（Logical Expressions）和空間推理（Spatial Reasoning），則非常困難。

Tasks involving computation, like math, require a "Chain of Thought" or extra computational resources during inference.
涉及數(shù)學(xué)這類需要大量計(jì)算的任務(wù)，需要使用“思維鏈”（Chain of Thought）或更多的計(jì)算資源。

Creative writing is particularly challenging because it involves world-building, storytelling, and maintaining plot coherence, making it easy for the model to lose coherence with just a small mistake.
創(chuàng)造性寫(xiě)作（Creative Writing）尤其困難，因?yàn)樗婕皹?gòu)建世界（World-building）、講故事（Storytelling）以及保持情節(jié)連貫性（Plot Coherence），模型很容易因?yàn)榧?xì)微的錯(cuò)誤而導(dǎo)致情節(jié)完全失去連貫性。

Evaluating creative writing is also difficult, making it one of the hardest open-ended AI research problems today.
創(chuàng)造性寫(xiě)作的評(píng)估也很困難，因此它是當(dāng)今最難的開(kāi)放式（Open-ended）AI 研究問(wèn)題之一。

From 2020 to 2021, the first major product based on scaling pre-training was GitHub Copilot, which used billions of code tokens from open-source projects.
從 2020 到 2021 年，基于擴(kuò)展預(yù)訓(xùn)練的首個(gè)主要產(chǎn)品是 GitHub Copilot，它使用了開(kāi)源項(xiàng)目中的數(shù)十億代碼 token。

Researchers improved its usability through Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning from AI Feedback (RLAIF).
研究人員通過(guò)人類反饋強(qiáng)化學(xué)習(xí)（Reinforcement Learning from Human Feedback，RLHF）和 AI 反饋強(qiáng)化學(xué)習(xí)（Reinforcement Learning from AI Feedback，RLAIF）提升了它的實(shí)用性。

This introduced a "post-training" phase focused on completing functions, generating multi-line completions, and predicting diffs.
這引入了一個(gè)“后訓(xùn)練”（Post-training）階段，專注于補(bǔ)全函數(shù)、生成多行代碼以及預(yù)測(cè)代碼差異（Diffs）。

The next major paradigm shift, published by OpenAI last year, is scaling reinforcement learning using "Chain of Thought" (CoT), enabling models to tackle highly complex reasoning tasks.
去年 OpenAI 提出了另一個(gè)重要范式轉(zhuǎn)變，即利用“思維鏈”（Chain of Thought，CoT）擴(kuò)展強(qiáng)化學(xué)習(xí)（Scaling Reinforcement Learning），使模型能夠處理高度復(fù)雜的推理任務(wù)。

In CoT, models spend significantly more computational time reasoning step-by-step through problems.
在思維鏈中，模型會(huì)花費(fèi)更多的計(jì)算時(shí)間逐步推理，解決問(wèn)題。

A major design challenge is how to present the model's complex thought processes to users without making them wait too long.
一個(gè)主要的設(shè)計(jì)挑戰(zhàn)是如何將模型復(fù)雜的思考過(guò)程呈現(xiàn)給用戶，同時(shí)避免用戶等待時(shí)間過(guò)長(zhǎng)。

This year is considered the "year of agents," characterized by complex reasoning, tool use, and long-context interactions.
今年被稱為“智能體之年”（Year of Agents），今年的特點(diǎn)是復(fù)雜推理（Complex Reasoning）、工具使用（Tool Use）和長(zhǎng)上下文（Long-context）的互動(dòng)。

The next stage will be agents evolving into co-innovators through creativity enabled by human-AI collaboration.
下一個(gè)階段，智能體將通過(guò)人類與 AI 的協(xié)作實(shí)現(xiàn)創(chuàng)造力，演變?yōu)楣餐瑒?chuàng)新者（Co-innovators）。

Future product research will involve rapidly iterating between highly complex models and smaller, distilled models.
未來(lái)的產(chǎn)品研究將涉及高復(fù)雜模型與更小、更快速的蒸餾模型（Distilled Models）之間的快速迭代。

Design challenges include making unfamiliar capabilities feel familiar (e.g., using file uploads), and enabling modular product features that scale easily.
設(shè)計(jì)挑戰(zhàn)包括使陌生的功能顯得熟悉（例如通過(guò)文件上傳）以及設(shè)計(jì)能夠輕松擴(kuò)展的模塊化產(chǎn)品功能。

Trust remains a key bottleneck; solutions include better collaborative interfaces allowing real-time user feedback and verification.
信任仍然是關(guān)鍵瓶頸；解決方案包括開(kāi)發(fā)更好的協(xié)作界面，使用戶能實(shí)時(shí)反饋和驗(yàn)證。

Innovative tools like Claude's Slack integration, ChatGPT tasks, and Canvas illustrate the potential of collaborative and multimodal AI interfaces.
創(chuàng)新工具，如 Claude 的 Slack 整合、ChatGPT 的任務(wù)功能和 Canvas，展示了協(xié)作與多模態(tài)（Multimodal）AI 接口的潛力。

Ultimately, the future involves "invisible software creation," allowing anyone, even without coding experience, to create and deploy tools through AI.
最終的未來(lái)愿景是“無(wú)形的軟件創(chuàng)造”，即使沒(méi)有編程經(jīng)驗(yàn)的人也能通過(guò) AI 創(chuàng)建和部署工具。

AI interfaces will evolve into highly personalized, multimodal, and interactive canvases, fundamentally changing how we interact with technology and the internet.
AI 界面將發(fā)展成高度個(gè)性化、多模態(tài)、互動(dòng)的“畫(huà)布”（Canvas），從根本上改變我們與技術(shù)和互聯(lián)網(wǎng)的交互方式。

“My prediction is that you will click less and less on internet links, and the way you will access the internet will be via model lenses, which will be much cleaner and in a much more personalized way.”

我預(yù)測(cè)，未來(lái)你在互聯(lián)網(wǎng)上的點(diǎn)擊量會(huì)越來(lái)越少；你訪問(wèn)網(wǎng)絡(luò)的方式將會(huì)通過(guò)“模型之鏡”完成，不僅界面更簡(jiǎn)潔，也更高度個(gè)性化。

“And you can imagine having very personalized multimodal outputs: let’s say if I say I want to learn more about the solar system, instead of it giving me a text output, it should give you a 3D interactive visualization of the solar system, and you can have highly rich interactive features to learn more.”

你可以想象這樣一種個(gè)性化的多模態(tài)體驗(yàn)：比如我想深入了解太陽(yáng)系，與其給我一段文字，不如呈現(xiàn)一個(gè)可交互的 3D 太陽(yáng)系可視化界面，并配備豐富的交互功能，幫助我更直觀、更深入地學(xué)習(xí)。

“I think there will be this sort of cool future of generative entertainment for people to learn and share new games with other people.”

我設(shè)想這樣一個(gè)很酷的未來(lái)：以“生成式娛樂(lè)”為媒介，人們不僅可以學(xué)習(xí)，還能隨時(shí)與他人一起創(chuàng)造并分享全新的游戲體驗(yàn)。

“I think the way I’m thinking about it is the kind of interface to AI is a blank canvas that kind of molds to your intent. So for example you come to the work today and your intention is to just write code, then the canvas becomes more of an IDE—like a cursor or like a coding IDE, although future programming might change.”

在我看來(lái)，與 AI 的交互界面就像一塊“空白畫(huà)布”，會(huì)根據(jù)你的意圖自動(dòng)定制。

如果你今天的目標(biāo)只是寫(xiě)代碼，這塊畫(huà)布就會(huì)變成一個(gè)類似 IDE 的開(kāi)發(fā)環(huán)境：自動(dòng)生成光標(biāo)、代碼補(bǔ)全、調(diào)試工具等（當(dāng)然，未來(lái)的編程模式也許會(huì)進(jìn)一步演進(jìn)）。

“Or if you’re a writer and you decided to write a novel together, the model can start creating tools on the fly for you such that it will be much easier for you to brainstorm or edit the writing or create character plots and visualize the structure of the plot itself.”

又或者你是一名作家，想和 AI 一起創(chuàng)作小說(shuō)，模型便會(huì)即時(shí)為你生成寫(xiě)作輔助工具，讓你更輕松地進(jìn)行頭腦風(fēng)暴、修改文稿、構(gòu)思角色線索，并可視化展示故事結(jié)構(gòu)。

“Finally, I think the co?innovation is actually going to happen with co?direction creative collaboration with the models itself, and it’s through collaboration with highly reasoning agent systems that will be extremely capable of superhuman tasks to create new novels, films, games, and essentially new science, new knowledge creation.”

最后，我相信“共同創(chuàng)新”將真正實(shí)現(xiàn)于人與模型的“共創(chuàng)共導(dǎo)”——通過(guò)與高度推理的智能體系統(tǒng)協(xié)作，這些系統(tǒng)將具備超越人類的能力，共同創(chuàng)作小說(shuō)、電影、游戲，乃至推動(dòng)全新的科學(xué)發(fā)現(xiàn)與知識(shí)創(chuàng)造。

“Cool. Um, thank you so much.”

太酷了。嗯，非常感謝大家！