本集简介
双语字幕
仅展示文本字幕，不包含中文音频；想边听边看，请使用 Bayt 播客 App。
当今人工智能研究的前沿正引领着什么方向？一些公司又如何已经在实际应用中运用这些技术？
Where is the cutting edge of AI research leading today, and how are some companies already putting it into action?
接下来，让我们与Cohere的首席人工智能官乔尔·皮诺一起探讨这个问题。
Let's talk about it with Cohere chief AI officer, Joel Pino, right after this.
本集节目由高通公司赞助播出。
This episode is brought to you by Qualcomm.
高通正在将智能计算带入每一个角落。
Qualcomm is bringing intelligent computing everywhere.
在每一个技术转折点上，高通都一直是值得信赖的合作伙伴，帮助世界应对最重要的挑战。
At every technological inflection point, Qualcomm has been a trusted partner helping the world tackle its most important challenges.
高通领先的AI技术、高性能低功耗计算以及无与伦比的连接解决方案，具备构建新生态系统、变革产业并改善我们体验世界方式的力量。
Qualcomm's leading edge AI, high performance, low power computing, and unrivaled connectivity solutions have the power to build new ecosystems, transform industries, and improve the way we all experience the world.
人工智能最有价值的应用是否在于工业领域？
Can AI's most valuable use be in the industrial setting?
在参观了IFS在纽约市举办的‘Industrial X unleashed’活动，并与IFS首席执行官马克·莫菲特交流后，我越来越频繁地思考这个问题。
I've been thinking about this question more and more after visiting IFS' Industrial X unleashed event in New York City and getting a chance to speak with IFS CEO, Mark Moffitt.
为了给出一个清晰的例子，莫菲特告诉我，IFS正在派遣波士顿动力的Spot机器人进行巡检，将数据传回IFS的神经中枢，再借助大型语言模型，为需要处理的区域指派合适的技术人员。
To give a clear example, Moffitt told me that IFS is sending Boston Dynamics spot robots out for inspection, bringing that data back to the IFS nerve center, which then with the assistance of large language models, can assign the right technician to examine areas that need attending.
这是技术的一个迷人前沿，我很感谢IFS的合作伙伴让我看到了这一点。
It's a fascinating frontier of the technology, and I'm thankful to my partners at IFS for opening my eyes to it.
如需了解更多，请访问 ifs.com。
To learn more, go to ifs.com.
那就是 ifs.com。
That's ifs.com.
欢迎收听《大科技》播客，这是一档针对科技界及其更广泛领域进行冷静而细致对话的节目。
Welcome to Big Technology Podcast, a show for cool headed and nuanced conversation of the tech world and beyond.
今天，我们将深入探讨人工智能研究的现状，前沿发展方向，当前方法是否存在局限性，以及一些公司如何在实践中实际应用这项技术。
Today, we're gonna look deep into the state of AI research, where the cutting edge is leading, whether there are limitations with the current methodologies, and how some companies are already putting this technology into action in a practical way.
我们非常荣幸地邀请到了完美的嘉宾——乔莉·皮诺在这里。
We're joined by the perfect guest, Joelle Pino is here.
她是Cohere的首席人工智能科学家。
She's the chief AI scientist at Cohere.
乔莉，欢迎来到节目。
Joelle, welcome to the show.
谢谢。
Thank you.
很高兴能来这里。
Glad to be here.
对于还不了解乔莉的人，她是一位在这行深耕已久的研究者。
So for those that don't know Joelle, she is a a, you know, a researcher who's been at this for a long time.
实际上，我和你是在ChatGPT发布后大约一个月见面的，那时人人都在问
You and I met, actually, maybe a month after ChatGPT was released and everybody was asking
嗯。
Mhmm.
AI是否具有意识。
Whether AI was sentient.
那时你是Meta基础人工智能研究部门的负责人。
You at that time were the head of the fundamental AI research division at Meta.
你同时也是麦吉尔大学的教授，目前担任Cohere的首席人工智能官。
You've you're also a professor at McGill, and currently, you're the chief AI officer at Cohere.
Cohere这家公司，我们之前邀请过艾丹·戈麦斯做客节目。
Cohere, we've had Aidan Gomez on the show.
他于2019年创立了这家公司。
He founded the company in 2019.
他也是《注意力机制就是你所需的一切》这篇论文的作者之一，这篇论文直接开启了生成式AI的时代。
He's also one of the authors of the attention is all you need paper, which basically kicked off the generative AI moment.
所以Cohere到现在已经有六七年历史了，给各位小朋友做个参考。
So Coherent is seven years old at this point, six, seven years old for the kids out there.
它已经融资了16亿美元。
It's raised 1,600,000,000.0.
它的估值达到70亿美元，主要向企业销售人工智能产品。
It's worth 7,000,000,000 and it sells AI to enterprise.
这为我们的讨论奠定了背景。
So that sets the stage.
是的。
Yes.
让我们聊聊人工智能研究。
Let's talk a little bit about AI research.
现在人们讨论得很多，关于人工智能研究是否会遇到瓶颈，以及这些新方法，比如在大语言模型上叠加强化学习、让模型进行推理、教模型使用不同工具。
There's so much discussion people have been talking about whether AI research is going to hit a wall and whether you know these new methodologies things like putting reinforcement learning on top of large language models going through reasoning, teaching the models to use different tools.
目前关于该聚焦哪里，有太多不同的观点。
There's so many different opinions of where to focus right now.
那么在你看来，人工智能研究的前沿是什么？你认为它将走向何方？
So what in your opinion is the cutting edge of AI research and where do you think it's going to lead?
我当然不担心研究会遇到瓶颈。
Well, I'm certainly not worried about research hitting a wall.
我们现在还有很多问题需要解决。
Like there's so many questions that we need to work on right now.
我可以将其分为两个有趣的视角，对吧？
And I'd separate it into two interesting angles, right?
一个是，现在究竟该解决哪些正确的问题？
One is like, what are the right problems to be solving right now?
当前的模型有哪些是做不到的？
What are the things that the models, the current generation of models we have can't do?
然后还有一个问题是，我们该如何着手？
And then there's a question of like how do we go about it?
什么样的假设能为我们解决这些问题提供线索？
What's the hypothesis that may give us the clue of how to solve some of these problems?
所以在该解决哪些问题方面，一个重要的问题是，我们该如何处理记忆？
So in terms of what problems to solve, think an important one is what do we do about memory?
机器有能力存储海量的信息。
Machines have the ability to remember tremendous amounts of information.
你只是把信息存进去而已。
You're just like stocking it in there.
困难的部分在于，知道何时该调用哪一部分信息来进行预测、生成内容或推理。
The hard part is knowing like when to pull on what piece of information to make a prediction, to generate information, to reason.
因此，能够更加精准地选择你所接触到的所有信息，这一点至关重要。
And so having this ability to be a lot more selective about all the information you've is seen in super important.
而Transformer模型已经是这一方向的重要一步。
And already transformers were an important piece of that.
你知道，注意力机制就足够了。
You know, attention is all you need.
但事实证明，光有注意力机制还不够。
Well, turns out it's not all you need.
你还需要更多东西。
You need a little bit more than that.
你需要具备在不同时间尺度和不同粒度上对信息进行推理的能力，等等。
You need the ability to reason about information at different timescales, at different granularity and so on and so forth.
因此，这方面还有很多值得深入研究的工作，这真正涉及到了架构选择、学习机制、数据集类型以及我们需要关注的使用场景。
So there's definitely a good piece of work to be done there, which really involves, and now we talk about the how, the choice of architecture, the choice of learning mechanisms, the type of data sets, type of use cases that we need to look into.
另一个重要的研究方向是构建世界模型。
Another big research theme is on building world models.
我们经常听到世界模型，它们本质上是吸收所有这些信息并预测行动效果的能力。
We hear a lot about world models, which are essentially the ability to take in all this information and predict the effect of actions.
当我们谈论因果性时，行动是如何改变世界的呢？
You know, so when we talk about causality, how are actions transforming the world?
这正是世界模型应该具备的能力。
This is what a world model should be able to do.
当你想要构建智能体时，世界模型是绝对必要的。
World models are absolutely essential when you want to build agents.
因为这些智能体会采取行动，从而改变世界。
Because these agents are going to take actions which is going to change the world.
你需要能够预测这些影响。
You want to be able to predict these effects.
所以，无论是构建机器人，讨论物理世界模型，还是部署在网上的智能体，
So whether you're building robots, and then we talk about physical world models, but also the agents getting deployed on the web.
你都需要构建数字世界模型，以便这些智能体——无论是做出财务决策、代表你沟通，还是安排会议——都能预测其行动的后果。
You need to build digital world models so that these agents, you know, whether they're making financial decisions, communicating on your behalf, organizing meetings, that they have the ability to predict the consequence of their actions.
所以这是一个重要的主题。
So that's a big theme.
关于如何构建这些世界模型，有许多不同的假设。
And there's a lot of different hypotheses about how to go about building these world models.
我要强调的第三个主题是——虽然还有很多其他主题，但至少让我选出最重要的三个——即如何高效地实现推理。
And the third theme I'll highlight, and there's many more, but let me at least pick out the top three choice, is about how do we build in reasoning efficiently.
目前，许多推理方法仍然依赖于较为彻底的前向搜索方法和学习正确的奖励函数。
And right now, lot of the reasoning methods are still quite thorough based on sort of forward search methods and learning the right reward function.
但我认为，推理、选择行动以及在不同粒度级别上进行规划，正迎来类似Transformer的突破时刻。
But I do think there's, you know, like the transformer moment for reasoning and choosing action and being able to plan at different levels of granularity.
我们离实现这一点还很遥远。
We're still far away from doing that.
因此，人们正在以各种方式将其融入系统，比如让大语言模型充当评判者，让AI系统之间相互提供反馈以进行训练。
And so there's all sorts of ways that it's being baked in, you know, LLM as a judge and things like that, where AI systems get feedback to AI systems in order to train them.
这仍然处于非常早期的阶段。
It's still very early days.
好的。
Okay.
我想深入探讨你刚才说的很多内容。
I want to dig into a lot of what you just said.
让我们从头开始。
Let's start at the beginning.
我们先从记忆说起。
Let's start with memory.
记忆和持续学习是同一枚硬币的两面吗？
Is memory and continual learning two sides of the same coin?
我的意思是，现在有这种观点，认为模型可以在会话中搜索网络并找到信息，但一旦关闭会话，它们就忘记了。
I mean, there's this idea that the models can search the web and they can find something in a session, but as soon as you close that session, they forget it.
是的。
Yeah.
我想我之所以提到这一点，是因为有些人提出解决这两个问题的方法是大幅扩大上下文窗口，然后更高效地在其中导航。
And I guess the reason why I'm going there is because a way that some people have suggested solving both of these is just making the context window massive and then just becoming efficient in the way that you navigate that.
是的。
Yeah.
你对这个假设有什么看法？
What do you think about that hypothesis?
这两个概念是相关的，但并不完全相同。
The two concepts are related, but they're not exactly the same.
记忆真正关注的是，在你试图解决的任务背景下，如何决定提取哪些信息。
And so memory is really about how do you address sort of what information to pull in, in the context of the task you're trying to solve.
持续学习假设上下文是不断变化的。
Continual learning makes the assumption that the context keeps on changing.
因此，你所学到的内容也在不断变化。
Therefore, what you've learned keeps on changing.
所以，非平稳性是持续学习的一个关键概念。
So, there's a notion of non stationarity that is really key to continual learning.
我承认，我对持续学习这个概念有些困惑，因为我感觉整个领域从未能清晰地界定这个问题，让大家达成一致。
I confess I have a little bit of trouble with continual learning as a concept because I feel the community has never been able to nail like how do we articulate the problem in a way that we all agree on it.
因此，每个从事持续学习工作的人对它的理解都不尽相同，这至少在我看来——虽然我在这个领域涉足不多——使得我们很难判断是否真的取得了进展。
And so everyone who does work on continual learning takes a different flavor of it, which makes it, at least in my eyes, and I haven't worked a lot in this area, but makes it a little bit hard to know whether we're making progress or not.
至于记忆，它的标准则稍微统一一些。
On memory, it's a little bit more standardized.
真正的张力在于效率与相关性之间的权衡。
The tension really is about, it's a question of efficiency and relevance.
因此，衡量你是否做得更好的方式相对更标准化，你并不想只是简单地记住一切。
So the way to measure whether you're doing that is a little bit better standardized and you don't want to be just sort of remembering everything.
所以，我们对任务的表述方式也相对更标准化了一些。
And so it's a little bit better standardized how we articulate the tasks.
好的。
Okay.
我们开始吧。
Let's go.
我们现在将讨论这两个方面，然后继续往下进行。
We're going to touch on both of those now and then we'll keep going down the list.
关于持续学习，也许我离这个领域太远了，我并没有遇到什么困难。
With continual learning, maybe I'm, you know, far removed from it, I'm not struggling with it.
所以我会给你我的原始想法，谈谈这是怎么回事，你可以帮我们稍微梳理一下。
So I'll give you my caveman thought about what this is and you can help us break it down a little bit.
人们已经指出这个问题：这些模型在运行过程中并不会发生变化。
I mean, the problem has been articulated that the models, they are, they don't change as they go about all these.
想象一下，如果GPT模型——每周与八亿甚至更多人交流——能够吸收这些对话，并从它所参与的讨论中学习，那该有多强大，虽然这可能有点吓人。
I mean, think about how powerful it would be if let's say the GPT model, which is speaking with 800,000,000 or maybe more by the time this comes out million people a week could, I mean, might be scary actually, but could internalize those conversations and learn from the discussions that it's having.
这几乎意味着，我同意你的观点，我们还没到瓶颈，但问题是：是否会有足够的数据来持续让这些机器变得更聪明？
That would almost, you know, you I I agree with you that the wall we're not at the wall, but the question is, is there going to be enough data to keep making these machines smarter?
当它们进行这些对话时，就打开了持续成长和学习的可能性。
And, you know, as they have these conversations, that opens up that ability to continue to grow and learn.
但无论它与人们进行多少次对话，模型本身却始终是静态的。
But the model stays static despite all the fact, all these conversations that it's having with people.
这不就是问题所在吗？而且
Isn't that the And
我的意思是，别误会，对吧？
I mean, don't get me wrong, right?
我真的相信我们必须解决这些模型需要持续进化的问题。
Like I absolutely believe we need to address the fact that these models need to keep on evolving.
对此我毫不怀疑。
I have no doubt about that.
我只是说，目前研究界在持续学习方面的进展，并没有真正与规模扩展方面的工作联系起来。
I just mean right now the progress in the research community that's working on continual learning isn't necessarily connecting to the work that's going on on scaling.
现在，发布的模型，你知道的，它们确实在持续进化。
Now, the models that are released, right, you know, they keep on evolving.
比如，我们今天拥有的生成式模型，无论是ChatGPT、Gemini，还是Cohere团队正在开发的Command模型，这些模型都在不断改进。
Would say, you know, the generative models we have today, whether it's ChadGPD, whether it's Gemini, whether it's the command models that the Cohere team is building, these models keep on improving.
只是我们并不一定让它们在线持续提升，而是选择在特定时间点发布，每次模型发布都有其特定的特性。
It's just, we don't necessarily let them improve online, but we ship at definite times, like a release of a model, has a particular characteristic.
这么做的优势在于，你可以在发布前充分测试模型。
The advantage of doing that, frankly, is you can really test the model before you put it out there.
可以全面测试它的性能、安全性等方面。
Can put it through its paces in terms of performance, in terms of safety and so on.
我会有点不愿意让模型自己持续运行，因为学习速度可能非常快，你可能会迅速脱离一个看似完全合理的状态，而这种情况过去我们已经见过几次了。
And I would be a little bit reluctant to just let the model keep running on its own because the learning can go very, very fast and you can switch out of a mode that seems completely reasonable very quickly, which we have seen a few times in the past.
是的。
Yeah.
我想我们可能在讨论同一个例子，就是微软那个叫Tay的机器人。
I think we might be thinking about one of the same instances when Microsoft had this bot called Tay.
对。
Yes.
我给你讲个故事。
I'll tell you a story.
我当年在Buzzfeed工作时，率先报道了微软有个叫Tay的机器人，我采访了相关人员，写了第一篇关于它的报道。
I actually broke the news of Tay that Microsoft was go had this great bot, spoke with the people, wrote the first story about it when I was at Buzzfeed.
我还把它置顶在我的Twitter个人资料上。
I pinned it to my Twitter profile.
我当时在西海岸睡觉。
I went to sleep on the West Coast.
嗯哼。
Mhmm.
我醒来时收到一堆消息，说你写过的那个聊天机器人——那个有趣的Dean机器人，竟然在宣扬纳粹意识形态。
And I woke up with all these messages being like, hey, that chatbot that you wrote about, the fun Dean chatbot is actually espousing Nazi ideology.
你可能得把那条推文取消置顶。
You might want to unpin that tweet.
这是因为它一直在学习。
And it was because it kept learning.
所以，好吧，也许可以继续学习，但前提是必须进行某种微调，以确保行为安全，也许应该是预防性微调。
So, okay, maybe continue learning, you know, if it's done because it has to also be done with some sort of fine tuning where you want to make sure that behavior maybe it's preemptive fine tuning.
那我们不如在实现持续测试之前，先别发布持续学习功能。
Well, let's not release continual learning till we've achieved continual testing.
这听起来是个非常合理的计划。
That sounds like a very reasonable plan.
好吧，关于记忆。
Alright, memory.
是什么让它如此困难？
What makes it so difficult?
我给你讲个故事。
I'll tell you one story.
我的周五搭档，周五搭档兰詹·罗伊和我，都试了谷歌的Gemini和Gmail。
My Friday co host, Friday co host, Ranjan Roy and I, we both went into Gemini on Google and on Gmail.
我们问：你能找到我发给我妻子的第一封邮件吗？
And we asked, can you find my the first email that I ever sent, with my wife?
好吧。
Okay.
它做不到。
Couldn't do it.
对。
Yep.
是因为邮件太多了吗？还是说，用AI去尝试找出你和妻子之间曾经进行过的对话本身就很难？
Is that because there's is it just because there's so many emails in there that actually like applying AI to try to figure out like what conversations have been had.
这很难吗？
Is that difficult?
还是说这是谷歌的产品问题？
Or is it kind of a product problem from Google?
为什么记忆功能这么难实现？我们最终会怎样解决它？
Like where, why is memory so difficult and how are we going to end up?
研究界将如何应对这个问题？
Like, how is the research community going to tackle this?
我的意思是，仅凭你的描述，要诊断这个问题有点困难。
I mean, like, it's a little bit difficult to diagnose just from your description.
我觉得自己就像一位外科医生，只能通过电话听病人描述症状。
I feel like I'm a little bit like, you know, a surgeon who's, you know, on the phone hearing the description of the patient.
你问过你的症状吗？
Have you asked about your symptoms?
所以我不会对你的具体情况做出精确的诊断。
So I won't necessarily venture a precise diagnosis for your case.
但即便如此，我认为弄清楚这个问题并不难。
But nonetheless, I don't think it's that difficult to figure out.
我的意思是，我得知道这个机器人是从哪里获取信息的，对吧？
Mean, I'd have to know what information is the bot pulling from, right?
比如在可见性和隐私方面。
Like, just in terms of visibility and privacy.
你有没有给它提供回答这个问题所需的所有信息？
Did you give it access to all of the information it needed to answer that?
这是第一个问题。
That's the first one.
我们实际上在Cohere做了很多部署，都是在客户现场进行的。
We do a lot, go back to what we're building at Cohere, actually we do a lot of deployments on-site.
有时候只是因为我们没有激活相关的访问权限来实现它。
Sometimes it's just a question, like we didn't activate the access information to do it.
所以，你需要判断是否获得了正确信息的访问权限，而且有很多原因让你不希望机器人随时访问你所有的信息。
So, you need to figure out whether that access to the right information is And there's all sorts of reason that you may not want to give the bots access to all of your information all the time.
所以，这是一个实际的考虑因素。
So, that's one practical consideration.
另一个是获取正确信息的问题。
The other one is like retrieving the right information.
那么，你的查询是否与信息的编码方式匹配呢？
And so, you know, did the query match how the information was encoded?
因为在这类系统中，你通常不希望直接以原始形式保存信息。
Because in most of these, you may not want to just leave the information in raw form.
那样成本会非常高。
It gets very expensive.
我的意思是，你个人可能无所谓，但考虑到一些公司运营的规模，你必须进行压缩，我们通常称之为嵌入。
I mean, you're one person, but at the scale that some of these companies are operating, you have to compress it, which we often called embedding.
所以你会创建这种表示的嵌入。
So you create like embeddings of this representation.
因此，它可能没有正确地将信息进行嵌入。
And so it may not have embedded the information properly.
然后是检索这些信息的问题，也许它检索到了一万条不同的内容，但正确的那一条却没排在前面。
And then there's like retrieving that information and maybe it retrieved like 10,000 different items and didn't drink this one close to the top.
所以它没有生成正确的回答，但可能它其实知道这条信息。
And so it didn't generate the right response, but it could be that it knows of it.
只是它没有出现在顶部。
It just didn't show up at the top.
因此有几种不同的原因，这让问题变得复杂。
So there's like a few different reasons, which makes it hard.
其中一个原因是信息访问的问题，当它对信息进行编码时，以及在正确时机检索信息时。
One of them is like the access to the information when it's encoding that information, then it's like retrieve the information at the right moment.
但当这些功能正常工作时，效果非常神奇。
But when this stuff works, it's pretty magical.
我刚才其实就在用Claude。
I was just in Claude, actually.
我注意到Claude的记忆能力有了很大提升。
And I noticed that Claude's memory capabilities have really improved.
我之前在和Claude对话时，喜欢上传我的访谈 transcripts，然后让它给我一些评分，比如从多个维度给我打个分。
Was speaking, so I love to upload the transcripts of my interviews and like, just, you know, get get a grading out of like, give me a rating on a variety of metrics.
你来决定。
You decide.
告诉机器人，你来决定。
Tight tell the bot, you decide.
你同意机器人给你的评分吗？
Do you agree with the ratings that the bot is giving you?
当然同意。
Definitely.
好的。
Okay.
通常来说，嗯，我确实这么做过。
Usually, well, I did.
有些不错，有些很差。
So some are good, some are bad.
我让Gemini做了一堆评分，结果所有类别都是505。
I actually had Gemini do a bunch of ratings and it was like five zero five on all categories.
我当时就觉得，这不对。
And I was like, that is wrong.
然后我去问了ChatGPT和Claude，它们的评价实际上合理多了。
And then I went to chat Chippy Tee and Claude and they were actually much more reasonable about it.
但本周我问Claude时，它做了一件有趣的事——开始把我这次的访谈和其他访谈做比较。
But one of the interesting things that Claude did when I asked it this week, it started comparing it to the other interviews I had done.
哦，原来如此。
Oh, okay.
它说，你知道，这次你确实抓住了更好的要点，这就是为什么我觉得上一次没那么有共鸣。
And it said, you know, you actually hit better points on this one and this is why this one didn't resonate in my opinion.
你有拿你的受众样本来做基准对比吗？
And did benchmark it with a sample of your audience?
这可能是下一步，我会从播客分析数据中提取信息并输入这些机器人。
That's probably the next and it'll probably when I, because I'll take data out of the podcast analytics and drop it in these bots.
它将能够进行交叉对比。
It's going to be able to cross reference.
所以当它正常工作时，简直神奇。
So when it works, it's magical.
你知道，你已经指出这是人工智能研究真正需要集中精力的领域之一，这正是前沿所在。
And, you know, you've identified this as one of the areas where AI research really needs to, you know, concentrate, and this is the cutting edge.
这种技术能发展到什么程度？
How good can this get?
你觉得呢？
And what do you think?
你认为现在正处于真正的突破时刻吗？
Do you think that it's at a moment of real progress?
还是说，让Claude做到我提到的那些事情，只是些花哨的演示技巧？
Or is it sort of party tricks to be able to get Claude to be able to do the things that I talked about?
关于评分的具体问题，比如分析和提炼一些反馈？
The question on ratings specifically, like analyzing and the sort of distilling some feedback?
更多是关于记忆，特别是它能够调用记忆这一点。
More about the memory, the fact that it can call back memory in particular.
不。
No.
我的意思是，我们在这一方面取得了不错的进展。
I mean, we're making good progress on that.
你知道，延长上下文长度是最简单的方法，但在这方面已经取得了相当多的进展。
You know, extending the context length is the kind of the easiest way to go about but there's quite a bit of progress that is that is being made on this.
好的。
Okay.
我们来谈谈推理。
Let's talk about reasoning.
你提到推理正处于一个前沿突破的时刻。
You mentioned reasoning as a as a cutting edge moment.
问题是效率。
The problem is efficiency.
这真的是问题所在吗？
Is that is that really the issue here?
我的意思是，推理就是模型一步步地进行。
I mean, so reasoning is the model basically goes step by step.
它尝试回答，检查答案，再尝试不同的答案，最终决定：好吧，这大概就是他们想要的。
It tries to answer, checks the answer, tries a different answer, then eventually decides, okay, this is probably what they want.
然后它输出结果。
And then it spits something out.
是的，我的意思是，大致就是这样的过程。
Yes, I mean, that roughly happens this way.
我认为真正的挑战在于能够以不同时间粒度进行规划，对吧？
I think the challenge is really being able to plan at different levels of sort of temporal granularity, right?
比如在执行行动时，假设你在计划一次旅行，对吧？
So in terms of how you execute actions, let's say, you know, you're, you're planning a trip, right?
你不会一开始就想着该穿哪双鞋去旅行，对吧？你会先从宏观层面考虑：大概什么季节、想去世界上的哪个地方，从最高层次开始。
You're not going to start by thinking of like, what are the shoes that I put on to go on my trip, right, you're going to start by talking thinking like, roughly what season roughly what, you know, part of the world do I want to go visit, you start from the top level.
然后你再往下细化一步，比如已经大致确定了时间和地点，接下来就要更精确地确定时间、地点、活动内容，以及可能和谁一起去。
And then you take it down a notch, which is like, okay, you've identified like a rough time, a rough place, like, let's get more precise on the time and the place in maybe the activity and like maybe who you want to go with.
然后再往下深入一层，对吧？
And then you take it down another notch, right?
这时候你才开始预订住宿和其他安排。
And that's when you start booking your reservations and so on.
但有时候，你会在预订环节遇到障碍，比如订不到想要的航班或酒店。
But sometimes, you know, you'll hit a blocker on the reservation and you can't get the flights or the hotel you want.
于是你会回退一步，问自己：要不要改日期？
And then you'll pop back up and say like, do I change my dates?
要不要换地方？
Do I change my place?
要不要换同行的人？
Do I change who I go with?
我不打算带孩子去，这样我们会有更多选择。
I'm not going bring the kids because then we can have more options.
所以，我们可以回到更高层次的分辨率上来。
So, we can pop back up in terms of level of resolution.
这部分是推理模型做不到的。
That's the part that the reasoning models don't do.
它们在单一粒度级别上表现得非常好。
They do really well at like one level of granularity.
比如你有一个机器人，给它所有手部和身体动作的指令，它就能规划并控制这些动作的细节。
So, you've got a robot, you give it all these like motions for the hands, the body motions, it can plan essentially to control the motors at that level of granularity.
但要在不同层次的行动分辨率之间来回切换，这非常困难。
But the going back and forth between different levels of resolution of action, it's really hard.
从技术上讲，我们称之为分层规划。
So on the technical terms, we call it hierarchical planning.
在来回切换时，要进行这种分解并保持信息的相关性，真的很难。
That's really hard to do that decomposition and keeping the information relevant as you go back and forth.
这是大型语言模型的局限性吗？
Is that just a limitation of the large language model?
因为LLM至少能够做到这一点，就像它最初是从预测开始的
Because the fact that an LLM can even do this in the first place, like again, like it started with predict
下一个词，它们在词级别上进行处理，
the next They do word have the word level,
对吧？
right?
没错。
Right.
从词级别出发，你确实能得到更高层次的内容。
And out of the word level, you do get the higher level.
这真的令人印象深刻。
It is really impressive.
我认为这才是让很多人感到震惊的部分。
I think that's the part that probably shocked a lot of people.
他们原本以为，比如在2023年，当你生成令牌时，是无法生成大型想法或更宏大的计划的。
They expected at, you know, back in twenty they '23 or expected that as you're generating tokens, you're not going to be able to generate sort of big ideas or a bigger plan.
然而，它确实做到了，这非常惊人，因此人们对它的看法出现了分歧：有些人觉得，嘿，这已经很了不起了。
Yet, it's pretty remarkable that it does it, which is why you get sort of different opinions in terms of some people's thinking like, Hey, it's already impressive.
让我们继续沿着这条路推进，最终会突破这个瓶颈；而另一些人则更加怀疑，认为你不可能实现这一点。
Let's just keep on pushing that way of doing things and we will unblock this and other people being a lot more skeptical that you'll achieve it.
能再解释得详细一点吗？
Explain that a little more.
所以，当它在打字时，我的意思是，我认为安德烈·卡帕西基本上解释过，Transformer就是一个计算机，每次生成一个新的令牌，你都在进行一次计算。
So as it's typing as it's, I mean, I think Andrej Karpathy basically explained that the transformer is a computer and every time you generate a new token, you're going through a piece of computing.
因此，你输入得越多，所使用的计算机就越大。
So the more you type, the bigger the computer is that you use.
是的，也就是说，输入的信息越多，你的表征就越庞大。
Yes, the more, I mean, more information goes in and the bigger your representation is.
明白了。
Okay.
所以，你是说随着这个过程发生，计算机实际上已经在提前思考了吗？
And so, but are you saying that as this happens, the computer is effectively already thinking ahead?
我举个例子。
I'll give one example.
Claude，回到Anthropic的一些研究，他们发布了一项令人惊叹的研究，让Claude写一首诗。
Claude, just to go back to some anthropic research, they published, this amazing research where they asked Claude to write a poem.
当它写下第一行时，模型中就已经激活了思考押韵的特征。
And as it's writing the first line, it's already activating features in the model that's thinking what rhymes with that.
是的。
Yeah.
这很了不起，因为这又是一种预测下一个词的技术。
Which is amazing because again, it's technology that predicts the next word.
下一个。
Next.
但当它预测下一个词时，已经在构思下一句话了，这对我来说简直不可思议。
But as it's predicting the next token, it's already thinking the next sentence, which to me is just mind boggling.
是的。
Yeah.
因此，这正是为什么在某种程度上，对代码的重视以及构建和生成代码表示的能力如此有趣，因为当你看代码时，对于有编程经验的人来说，代码具有那种结构，那种层次结构，是被编码在内的。
And so, I mean, this is why to some degree, the emphasis on code and the ability to build representations of code and generate code is so interesting because when you look at code, and for people who have programmed before, the code has that structure, that hierarchical structure, it's encoded in.
任何看过一堆代码的人，即使不是某种特定语言，也能理解函数、变量、库等概念。
Anyone who looks at a bunch of code, even if it's not necessarily a language, understand the notion of functions and variables and libraries and so on.
因此，项目的这些不同粒度层次，都编码在其中。
And so, those different levels of granularity of the project, it's encoded in there.
因此，有人希望，通过大量代码训练，机器本质上能推断出这些结构线索。
And so, there's some hope that by training enough on code, the machine essentially infers these kinds of structural cues.
令人着迷。
Fascinating.
所以，就像你提到的，这项技术的这一特性——这正是让我有点难以置信的地方。
So, like you talked about, that the the fact that this technology, and this is sort of the thing that that sort of makes my head explode a little bit.
这项技术能够做到这些事情，而按照其架构，你本不会认为它能做到这些。
The fact that this technology is able to do these things that you wouldn't think given the architecture it is supposed to do.
同样的道理也适用于视频模型和图像模型。
Same with if you think about video models and image models.
顺便说一下，你的一位前同事杨立昆总是谈论如何生成视频，我知道他对视频模型有些批评，但要生成AI视频，你真的必须能够预测和规划接下来会发生什么。
And by the way, one of your former colleagues, Jan Lecun, would always talk about how to generate and I know he has some criticisms of video models, but to be able to generate AI video, you really have to be able to predict and plan what's gonna happen
在物理世界中。
in the physical world.
没错。
Absolutely.
甚至在顶尖研究人员中，我认为也没有完全理解这一点：当你让模型做，比如杨立昆最喜欢的例子——让一支铅笔掉下来时。
And there's some embedded intelligence that even leading researchers, I don't think fully get that when you, for instance, ask a model just to use Jan's favorite example, to drop a pencil.
这其中可能的变数太多了。
There's so many permutations of where that can go.
而现在，这些模型即使没有接受过物理课程的训练，也能理解铅笔会下落，可能会撞到桌子，甚至可能弹起来。
And now the models without like, I mean, without having lessons of physics, understand that it drops and maybe hits the table and might bounce up.
是的。
Yep.
因为模型已经见过足够多被掉落物体的数据，这些物体都表现出类似的行为。
Because it's seen enough data from objects that are dropped that have these kinds of behavior.
但试着预测一个类似的物体在不同星球上被掉落时的行为。
But try to predict what's the behavior of a similar object dropped on a different planet.
而预测很可能是错误的，因为所有数据都是在我们地球的重力常数下采集的。
And probably the prediction is wrong because all of the data was taken with our gravity constant.
是的，说到这个，我刚刚看到一个生成的视频，画面中一个人拿着泡沫塑料杯时，手指竟然从杯子里伸了出来。
Yes, I mean, will say as I'm talking about this, I did just see a video generated where a man's fingers came out of a Styrofoam cup as he was holding it.
还有很多改进空间。
There's room to still a lot of room for improvement.
最近有一些讨论，德米斯·哈萨比斯谈到谷歌的视频模型在某种程度上具备了类似世界模型的能力。
Now there is some talk, Demesis Abbas was on recently talking about how Google's video models in some way have capability like these world model capabilities.
它们确实理解物理规律，而你提到的世界模型是这项技术仍有巨大潜力发展的另一个领域，目前仍不太明确。
They do understand the physics and you brought up world models as another area where this technology really has the potential to grow its cutting edge, still kind of undefined.
我想说的是，回到原始人的话题，我对为什么例如你之前提到的一个例子感到有些困惑：如果你想让模型能够外出完成金融交易并理解金融交易的含义，它就必须了解世界是如何运作的。
I will say, you know, going back to the caveman here, I'm a little bit confused about why, for instance, like one of the examples that you brought up earlier was that if you want a model to be able to like go out and like complete financial transactions and understand the implications of financial transactions, it has to know how the world works.
是的。
Yep.
但你不能通过文本来教授这一点吗？
But can't you just teach that in text?
你不能通过文本或者数字逻辑来教它吗？比如，如果你用我的信用卡在线购物，我会破产，所以别这么做。我认为世界模型就像你所说的，这些模型需要理解重力。
Can't you teach it like if you use my credit card, and you know, buy anything online, I will go bankrupt, like in, in text or even number logic, and therefore don't do it like, why does and I think world models is like you, these models need to understand gravity.
为什么模型需要理解重力，才能学会这些关于世界运行的基本规则？
Why does a model need to understand gravity to learn these basic, rules of sort of the way that the world works?
嗯，这正是我之前区分物理世界模型和数字世界模型的原因，对吧？
Well, and this is why earlier I sort of distinguish between like physical world models and digital world models, right?
有可能你确实可以构建出非常有效的基于网络的代理，而它们根本不需要理解重力的概念。
It's it's possible that you can actually build really effective agents, web based agents that don't understand the concept of gravity.
也有可能你能够构建用于机器人的物理世界模型，而它们根本不需要理解银行系统的运作方式。
And it's possible you can build physical world models for robots that don't need to understand, you know, the functioning banking system.
因此，你可以将“世界”定义为一个封闭的环境。
And so you can define the word world as being like a contained environment.
因此，但如果你希望在该环境中部署代理，那么它就必须很好地理解该环境的规则。
And so, but if you want to deploy the agent on that environment, then it does need to understand the rules of that that environment quite well.
挑战在于获取足够覆盖所有可能未来场景的数据，也就是在各种事件发生下，世界可能演变的种种方式。
The challenge is getting enough coverage of data for all the possible futures, right, and all the different ways that the world could evolve subject to various events happening.
因此，如今最有益的应用场景往往是那些需要人类参与其中的情况。
So, a lot of the cases today where it's actually most beneficial is where there's like a place for the human at the table.
我给你举个例子，好吗？
And I'll give you an example, right?
人们经常谈论用聊天机器人提供客户服务，对吧？
People talk a lot about using chatbots for customer service, right?
比如，聊天机器人应该直接接入系统，就能回答所有问题，24/7随时在线等等。
Like chatbots should be, you should just plug them in, they will answer all your questions, they'll be available 20 fourseven and so on.
但实际上，当然会有很多聊天机器人被部署在这些场景中，但我们看到的一个非常有效的用例是让机器人整合所有相关信息。
In reality, and there will be of course, many chatbots deployed for these kinds of cases, but you know, like one of the use cases we've seen that works really well is actually to have the bot pull together all the relevant information.
在客户服务中，机器人从多个不同来源整合所有相关信息，而不是遵循固定脚本只是个聊天机器人——而是把关于系统文档、客户案例、各种问题描述等所有信息都汇集起来，然后提出诊断结果和若干建议操作，并保持人类在环路中，以验证方案并执行操作。
Due to customer service, you pull together all the relevant information from many different sources, as opposed to following a script and being just chat bot, pull together all that information about, you know, the documents, the documentation that accompanies the system, the case on the client, the different problem, that description that you have, pull all together that, and then you pose a diagnostic, and then you pose a few suggested actions, and then you keep a human in the loop to validate the plan and to carry out the action.
这意味着人类可以参与进来，而这些情况比你的手机套餐之类的问题要复杂得多。
And so that means that the human, you know, can and these are more complicated cases than just like your cell phone plan or something like that.
但即便如此，在这些情况下，原本可能需要半小时才能整合并提炼出所有信息，现在你可以将这个过程缩短到二十秒左右：分析、验证并执行操作。
But nonetheless, in those cases, what would have taken a long time, maybe half an hour to pull together all that information, distill it for a human, now you can reduce that down to a twenty second you know, analyze, verify and carry out the action.
因此，如果你能将人类与AI代理结合起来，往往能获得更强大的结果。
So if you have that ability to combine the human and the AI agent, actually you get often some much more powerful results.
这意味着，如果你的世界模型不够完整，人类可以补上缺失的部分，提供额外信息，然后将这些信息反馈回来训练你的代理，从而实现持续学习。
And it means if your world model isn't complete, humans in the loop, they figure out the pieces that's missing, they give that extra information, and then you bring that information back to train your agent, then you get continual learning.
就是这样。
There you go.
我们正在逐步接近。
We're getting there.
你认为要实现通用人工智能，模型必须理解重力吗？
Do you buy that the models need to understand gravity for AGI to be reached?
我的意思是，目前主要有两种观点：一种认为你可以仅通过比特、字母、图像等数据来训练通用人工智能；另一种观点则认为，你真的需要让模型理解的不只是扑克的规则，而是当一个人把手放在扑克桌上时会发生什么。
I mean, there are basically like a couple schools of thought that you could you could basically train AGI on on bits and, you know, letters and stuff like that, images, and then there are others that believe, you really need these models to understand, like, not just the rules of poker, but like what happens when a person puts their hand on a poker table?
你觉得呢？
What do you think?
是的，我的观点是，与其指望我们会诞生一个单一的超级智能代理，不如说我们更可能生活在一个拥有众多专用代理的未来。
Yeah, I mean, I tend to actually place my bet, not on the fact that we're gonna reach like a single super intelligent agent, but on the fact that we are much more likely to live in a future where there's going to be many agents for many things.
因此，某些代理确实需要理解重力。
And so, some agents will absolutely need to understand gravity.
如果你要让物理机器人在世界上移动、碰撞物体、拾取物品等，它们就必须理解这一点。
You know, if we're gonna have physical robots that are moving around in the world, that are going to be hitting objects, that are going be picking up objects and so on, they will need to understand that.
而其他处理我们数字生活的代理则可能不需要理解这一点。
Other agents that are dealing, for example, with our digital life may not need to understand that.
我们还需要为这些代理建立一种交互和沟通的协议。
And we also need to have a protocol for these agents to interact with each other and to talk to each other.
因此，我认为更有可能的情况是，而不是让一个无所不知、拥有完整世界模型的超级代理来掌控一切。
So I actually think that's a much more likely scenario, rather than have the uber agent that needs to understand everything and have a fully encapsulated world model.
最近，AI实验室的领导者们一直在说一个流行的观点。
There's a popular thing that AI lab leaders have been saying recently.
他们一直在谈论能力过剩的问题，即人工智能技术能做的远比当前实际应用的多。
They've been talking about how there's a capability overhang, how AI technology can do a lot more than it's being used for.
你相信这一点吗？
Do you believe that?
当然。
Absolutely.
是的。
Yeah.
再多说说这个。
Say say more about it.
你觉得有哪些本可以做却还没做的事情？
Talk about what do you think is not being done that could be done?
我每天都能看到。
I see it every day.
我的意思是，这打开了一点小窗口。
I mean, it all open up a little window.
我之所以对加入Cohere感到非常兴奋，原因之一是它是为数不多的拥有研究团队的公司之一。
Like one of the reasons I was super excited about joining Cohere is because it's one of the few places that you know, that we have a team that does research.
所以我每天都能看到研究领域正在发生什么。
So I get to see, you know, day to day what's happening in research.
我们有一个做建模的团队。
We have a team that does modeling.
所以我能看到我们正在构建的模型，查看评估结果，以及完整的评估范围。
So I get to see the models that we're building, look at the evaluations, the full spread of evaluation.
而且我们还有一个产品。
And we have a product.
这个产品是一个面向真实客户的智能代理平台。
That product is an agentic platform that is going to real clients.
所以你能看到整个全貌。
So you get to see the whole thing.
我看到我们的模型能做什么，也看到我们产品中实现的功能，然后发现很多客户因为各种原因没有充分利用所有功能。
And I see something that our models can do, and I see some things that we've built into the product, and then we go and there's a lot of customers that are not using the full functionality for all sorts of reasons.
展开剩余字幕（还有 307 条）
所以，我认为我们在能力方面和当前实际部署的内容之间存在巨大差距。
So, think that between what we have in terms of capacity versus what's being deployed right now, there's a big gap between that.
有时原因在于容量问题。
Sometimes the reasons are capacity questions.
我们经常谈论超级智能和大模型。
We talk a lot about superintelligence big models.
但实际上，付费客户更看重性能与效率之间的良好平衡。
In reality, paying customers want a good trade off in terms of performance for efficiency.
因此，我们会训练更大的模型，但部署更小的模型，因为这样能实现更好的平衡。
So, we'll train bigger models, but we'll deploy smaller models because it gives us that trade off.
这是一种足以完成任务的足够智能。
It's like good enough intelligence to get the job done.
我会说，我们可以提供给你更多功能。
And I'm like, well, we could give you so much more.
不，这样已经足够了。
No, that's good enough.
所以，对他们来说，采取这种立场完全是合理的。
So and it's a perfectly, you know, rational position for them to be taking.
因此，其中一部分是出于效率的考虑。
So some of that is for efficiency reasons.
另一部分差距是因为你正在进入那些已有既定系统和流程的组织。
Some of that gap is also because you're going into organizations which have systems and processes in place.
有时候，这些流程当前的设计目标与一个更欢迎AI代理的环境之间存在不匹配。
And sometimes there's like a mismatch between what those processes are set up to do today versus what would be a more welcoming environment for an AI agent.
所以，存在这类问题。
So, there's these kinds of things.
另一个方面是，我认为很多知识并没有被编码化。
And then the other one is often I think there's a lot of intelligence that is not encoded.
因此，这些代理会接入各种内部系统，在考虑隐私和安全的前提下，利用所有的业务智能。
So, the agents go, they plug into a bunch of internal systems, they leverage all the business intelligence with privacy security consideration.
它们利用了所有这些信息，但有时仍存在大量目前尚未被利用的信息盲区。
They leverage all that information, but sometimes there's big pockets of information that we're not leveraging right now.
如果我们能接入这些，就能做更多事情。
And if we did, if we connected into that, then we would be able to do a lot more.
因此，从组织或个人向人工智能传递信息时存在的这种不匹配，是导致大量机器智能被浪费的另一个原因。
So that impedance mismatch in terms of the information sharing from the organization or from the individual to the AI is another case where it leaves a lot of machine intelligence on the table.
我们一会儿会讨论企业领域，但我先问你一个问题：这在消费领域如何应用？
So we're going to talk about enterprise in a moment, let me ask you one question about how this applies to consumer.
显然，大家已经讨论了很多技术，大型科技公司都有一个愿景，那就是打造一个通用助手，比如苹果智能或升级版的Alexa，它们都已经以各自的方式推出了。
Obviously, talked about a lot of technology and the vision is there within the big tech companies to have a like a universal assistant, something like an Apple intelligence or an Alexa plus, you know, both of them have rolled out in their own way.
但它们——还有Meta、谷歌各自的产品——都没有引起轰动。
But both both of them and I guess, know, Meta has their own product, Google has their own product.
这些产品都没有让世界为之沸腾。
None of these are lighting the world on fire.
你认为这是为什么？
Is do you think that is?
这是否又是能力过剩的另一个例子？
Is this another example of a capability overing?
还是说技术本身还没到位？
Or is it that the technology is just not there yet?
我觉得两者都对。
I think both are true.
我觉得，人们期待的是， basically 被承诺了超级智能。
I think, you know, people are expecting, you know, basically been promised superintelligence.
所以他们期待这些AI系统能带来奇迹，但其实这并不是奇迹。
So you know, they are expecting magic out of these, these AI systems, It is not magic.
因此，我认为当前人们对它们能力的期望与实际能力之间存在巨大差距。
And so I would say there's a big gap between expectation of what they can do today.
同时，人们试图做的事情与这些智能体真正擅长的方面之间也存在不匹配。
And then there's also a mismatch between what people try to do versus what might be the strength of these agents.
我打个比方，你在一个团队里，新来的同事第一天加入，你可能还不清楚他能做什么、不能做什么，需要一段时间共事才能了解。
I compare it a little bit, you you're working in a team, you get a new teammate in like day one, you may not know exactly what this person is capable of, not capable of, and it takes some time working together.
有时候，当你给他更多信息时，这个人会表现得更好。
And sometimes that person gets a lot better when you give them a lot more information.
有时候你会发现他们拥有了之前没有的新技能。
And sometimes you discover they have a new skill that they didn't have.
但归根结底，你知道，这个人往往无法随时随地完成所有事情。
But at the end of the day, you know, often that person isn't able to do everything everywhere all at once.
所以我认为，这两点其实同时都是对的。
And so I think there's, there's both both these things are true at the same time.
是的，还有大量的企业政治因素，我最近刚写了这个
Yeah, there's also I mean, lot of corporate politics, I just wrote this
当然，
course,
我最近在《科技巨头》上写了一篇故事，讲的是其实有两种基本路径，而你正好非常适合谈论这个话题，或者给我们讲讲真实情况。
I wrote this story recently in big technology, talking about how there's like these two basic and actually, you're in a great position to talk about this or give us the real story here.
从我的角度来看，很多公司基本上正沿着两条路径前进。
From my vantage point, there's basically two trajectories that a lot of companies are on.
公司本身并没有在跟客户谈论这个，但如果你从整体上看这些公司，很多公司都难以将这项技术落地，但个人已经开始感受到它的益处。
The companies themselves have not talking about your customers, but if you take about like companies overall, many of them have struggled to put this technology into place, but individuals are starting to see the benefit.
所以你实际上看到一些公司正在进行试点项目，但这些项目并没有进入生产阶段，而与此同时，可能有些基层员工正在使用Claude Code，并且真的从中获益。对此你怎么看？
So you actually have like these companies with these pilots that are not getting into production, but then you might have somebody, you know, lower down using Claude Code, who's like actually getting So this what do you what do you think about that?
如果我们看到这种分化持续下去，你认为这意味着什么？
And what do you think it means if we end up seeing that divergence continue?
我认为这完全正确。
I think that is absolutely true.
我经常看到这种情况，甚至在我们自己的公司里也是如此。
See this all the time, even within our own companies.
人们利用这项技术的能力差异很大。
People's ability to leverage the technology varies a lot.
事实上，我们正朝着一个越来越多地使用这项技术的世界迈进，因此，那些能够理解和利用这项技术的人将占据优势。
I mean, the reality is we are moving towards a world where there's going to be more and more of that technology, and so the people who have the ability to understand and leverage the technology are going to have an edge.
好的。
Okay.
我同意。
I agree.
好的。
All right.
在我们休息一下，转到更多应用和更连贯的内容之前，最后一个问题是：
Last question before we take a break and go on to some more of the applications, some of the more coherent stuff.
我仍然难以理解，为什么这些AI实验室在技术成果上如此接近。
I still can't wrap my head around the fact that the AI labs are so close together in terms of the technology they produce.
一个实验室做出了创新，下一个实验室就立刻拥有了同样的创新。
One build some innovation, the next has the innovation.
一个似乎突然领先了。
One seems like it leaps ahead.
下一个又似乎突然领先了。
The next seems like it leaps ahead.
你能想象出某个实验室突然发现某项突破，从而真正拉开与其他实验室差距的情景吗？
Can you envision a scenario where one of the labs just like kind of hits on something and can actually open up a lead against the others?
还是说它们将永远并驾齐驱？
Or is it just going to be neck and neck forever?
我认为很难把想法局限在某个范围内。
I think it's really hard to keep ideas in a box.
尤其是因为这些想法往往存在于人们的脑海中。
Especially because in many ways, these ideas they reside in, in people's heads.
你和我一样，都见过人们在这些公司之间流动。
And you've seen as much as me the movement of people between these companies.
他们总是来回跳动，带着这些想法一起走，即使代码仍留在原地。
They're always ping ponking back They in carry the ideas with them, even if the code stays on one side.
一旦你看到了某种洞见，就再也无法视而不见。
Once you've seen some insight, you can't unsee it.
对。
Right.
所以，他们可能需要重新实现，可能需要用不同的方式表达，甚至给它起个不同的名字，但想法总会传播开来。
And so, you know, they may need to re implement, they may need to articulate in different ways, they may give it a different name, but ideas just circulate.
你无法把想法局限在某个范围内。
You can't keep ideas in a box.
因此，老实说，多年来我一直大力倡导开放科学。
That's why honestly, for many years I've been so much an advocate for open science.
我只是不相信，除非你愿意把人也关起来，否则你无法把这些想法封锁住——而我们并不愿意这么做。
I just don't believe that you can keep these ideas boxed in unless you're willing to keep people boxed in, which we are not willing to do.
所以，我认为我们有一种方式可以封闭这些想法。
And so, I think we have a way to close the ideas.
我们应该接受这样一个事实：当让想法自由流动时，我们所有人都能更快地进步。
We should embrace the fact that when you let the ideas circulate, all of us progress faster.
那么问题来了，假设所有这些实验室都达到了超级智能。
And then the question is, let's say all these labs do reach super intelligence.
人们一直问，你不可能垄断它。
You know, it's been asked, well, you can't hoard it.
那么，开发它有什么经济价值呢？
So where's the economic value in developing it?
是的。
Yeah.
我们还处于这项技术的非常非常早期阶段，甚至在人工智能时代，什么是主导的经济模式、什么是正确的商业策略方面，我们更早得几乎一无所知。
We're still very, very early days in the technology and we're even earlier days in terms of like what are going to be the dominant economic models, what is going to be the right business strategy in the age of AI.
我认为我们需要给自己时间去尝试。
And I think we need to give ourselves the time to experiment.
现在我们已经有大约三十年的视角来看待互联网及其经济影响，而要弄清楚这一点也需要好几年的时间。
Know, now we have thirty years or so perspective on the internet and the economic impact of that, and it's going to take a number of years before we figure that out.
但通常，开发技术的人并不一定就是扩大技术规模的人，也不一定是将其商业化的人，更不一定是对它进行控制和监管的人。
But often, you know, those who develop the technology are not necessarily the same as those who scale the technology versus those who actually commercialise it, versus those who actually control it and regulate it.
因此，将由此产生一个相当复杂的生态系统。
So there's a pretty complex ecosystem that is all going to arise out of that.
好的。
Okay.
在休息之后，我们将讨论这项技术已经产生的实际经济影响。
Well, at the other side of this break, we're going talk about some real economic impact of this technology already.
稍微谈一谈Cohere正在做什么，然后我们会深入更多内容。
Talk a little bit about what Cohere is up to, and then we'll cover a lot more.
我们马上回来。
So we'll be back right after this.
你想吃得更好，但你一点时间都没有，也没精力去实现。
You wanna eat better, but you have zero time and zero energy to make it happen.
Factor 不要求你提前备餐或遵循食谱。
Factor doesn't ask you to meal prep or follow recipes.
它只是彻底解决了这个问题。
It just removes the entire problem.
两分钟，真材实料，搞定。
Two minutes, real food, done.
还记得那次你想做健康餐，却没时间做的时候吗？
Remember that time where you wanted to cook healthy but just ran out of time to do it?
你并不是在健康饮食上失败了。
You're not failing at healthy eating.
你只是缺少额外的三个小时。
You're failing at having an extra three hours.
Factor的餐食由厨师精心制作，营养师设计，并直接配送到您家门口。
Factor is already made by chefs, designed by dietitians, and delivered to your door.
您只需加热两分钟即可享用。
You heat it for two minutes and eat.
餐内包含优质蛋白质、色彩丰富的蔬菜、全食物原料和健康脂肪——这些都是您如果有时间会自己做的食材。
Inside, there are lean proteins, colorful vegetables, whole food ingredients, healthy fats, the stuff you'd make if you had the time.
此外，我们还推出了全新的肌肉强化系列，专为增肌和恢复设计。
There's also a new muscle pro collection for strength and recovery.
您始终能享用新鲜的餐食。
You always get to eat fresh.
只需两分钟即可准备好。
It's ready in two minutes.
无需准备，无需清理，无需精神负担。
No prep, no cleanup, no mental load.
前往 factormeals.com/bigtech50off，使用代码 big tech 50 off，即可享受首单 Factor 餐盒 50% 折扣，并免费获得一年早餐。
Head to factormeals.com/bigtech50off and use code big tech 50 off to get 50% off your first factor box plus free breakfast for one year.
此优惠仅适用于使用优惠码并购买符合条件的自动续订订阅的新Factor客户。
Offer only valid for new factor customers with code and qualifying auto renewing subscription purchase.
用Factor让健康饮食变得更简单。
Make healthier eating easy with factor.
开始一件事并不只是困难。
Starting something new isn't just hard.
它令人恐惧。
It's terrifying.
为此付出了这么多努力，你却不确定它是否能成功，要迈出这一步确实很难。
So much work goes into this thing that you're not entirely sure will work out, and it can be hard to make that leap of faith.
当我刚开始做这个播客时，我不确定是否有人会听。
When I started this podcast, I wasn't sure if anyone would listen.
现在我知道，这是个正确的选择。
Now I know it was the right choice.
当你有Shopify这样的合作伙伴在你身边提供帮助时，也会更有帮助。
It also helps when you have a partner like Shopify on your side to help.
Shopify 是全球数百万企业的电商平台，占美国所有电子商务的10%。
Shopify is the commerce platform behind millions of businesses around the world and 10% of all ecommerce in The US.
从Allbirds和Cotopaxi这样的知名品牌，到刚刚起步的新锐品牌，都在使用它。
From household names like Allbirds and Cotopaxi to brands just getting started.
凭借数百个即用型模板，Shopify 帮助你打造一个与品牌风格完美契合的精美网店。
With hundreds of ready to use templates, Shopify helps you build a beautiful online store that matches your brand style.
你还可以像拥有整个营销团队一样轻松推广，随时随地为在社交媒体或网络上浏览的客户创建电子邮件和社交媒体活动。
You can also get the word out like you have a marketing team behind you, easily create email and social media campaigns wherever your customers are scrolling or strolling.
现在是时候用 Shopify 将那些‘如果’变成现实了。
It's time to turn those what ifs into with Shopify today.
立即前往 shopify.com/bigtech 注册每月1美元的试用。
Sign up for your $1 per month trial at shopify.com/bigtech.
访问 shopify.com/bigtech。
Go to shopify.com/bigtech.
就是 shopify.com/bigtech。
That's shopify.com/bigtech.
我们回到《大科技》播客，今天邀请到Cohere公司的首席人工智能官乔尔·皮诺。
And we're back here on big technology podcast with Joel Pino, the chief AI officer at Cohere.
当然，这也是我们在达沃斯高通之家举办的达沃斯系列访谈，并将在接下来几周持续进行。
And, of course, this is part of our Davos series that we're hosting at the Qualcomm house here in Davos and running over the weeks following.
所以，乔尔，很高兴你能来。
So, Joel, it's great to have you.
我来谈谈我所了解到的AI在商业中的应用场景，你看看我有没有遗漏，然后说说你认为最有价值的是哪一个。
Let me give you what I've gathered as the use cases in business for AI, and you tell me if I'm missing any, and then maybe what you think is the most valuable.
好的。
Alright.
我列了四个。
I wrote four down.
第一个是外部聊天机器人，也就是客户互动型聊天机器人，就像布雷特·泰勒在Sierra提到的那种。
One is external chatbots, the customer engagement type of chatbots, the type like Brett Taylor talked about at Sierra.
另一个是内部知识。
The other is internal knowledge.
假设一家公司内部有知识，但这些知识是零散的，你可以通过一个机器人来查询这些内部信息。
So let's say a company has knowledge within the company and it's all fragmented and maybe there's a bot that you can start to query internal knowledge.
第三是用AI弥补那些不顺畅的系统。
Third is papering over systems that don't work.
我觉得这不需要更多解释了。
I don't think that needs much more explanation.
这就像一个故事
It's like the story
我对这一点持怀疑态度，但还是。
I'm of skeptical about that, but still.
第四是自动化。
And then the fourth is automation.
是的。
Yeah.
在企业AI的应用中，我有没有遗漏什么重要类别？你认为目前真正有价值或最大的类别是什么？
Am I missing any big categories in as far as AI in business and where do you think the real value or the biggest category is right now?
我认为可以从不同角度来划分它。
I think there's like different ways to slice it.
我认为这是一种非常合理的划分方式。
I think that's a perfectly reasonable way to slice it.
我见过的另一种划分方式是将AI分为预测性AI、生成式AI和代理式AI，后者是完全不同的另一个层次的机会。
I think another way that I've seen it sliced is between like predictive AI, generative AI versus agentic AI, which is like a whole other level of opportunity.
另一种划分方式则是按应用领域来分。
And then the other way I've seen it sliced is more by application domains.
比如AI在医疗领域会做什么，AI在科学发现中会发挥什么作用，AI在银行业会怎样应用，以及在公共部门等领域的应用。
Whether it's what AI is going to do in healthcare, what AI is going to do for scientific discovery, what it's going to do in banking, what it's doing for example public sector and so on.
这是人们看待不同机会类别的一种方式。
That's the other way that people have looked at the different classes of opportunity.
那你认为最大的是哪一个？
And so what do think the biggest is?
潜力巨大。
There is so much potential.
我有点犹豫不敢选一个。
I hesitate to pick one.
我会说，坦率来讲，Cohere押注的方向以及我们的核心假设是企业级AI，它需要极高的隐私和安全保证。
I will say, you know, quite frankly, where Cohere has placed its chips and the core hypothesis is on the case of enterprise AI that needs really high privacy and security guarantees.
我认为有一大类应用属于你提到的第二类，那就是你拥有大量内部业务情报信息，但这些信息可能是分散的。
I think there's a big cluster of applications which falls a little bit in the second category that you outlined, where you know, you have a lot of internal business intelligence information, perhaps fragmented.
你希望利用所有这些信息来赋能你的员工。
You want to be able to leverage all that information to empower your employees.
因此，当这些信息是你不希望通过API泄露到网络上的时候，就存在一个机会，可以构建在内部运行、基于本地数据的智能系统，帮助员工工作，本质上成为员工的亲密伙伴。
And so, that case, especially when that information is something that you don't want to pop up on the web through an API, there's an opportunity to build the genetic systems that work in house with the local data that inform the employees and are essentially like close partners to the employees.
你能给我一个具体的应用案例或案例研究吗？
Can you give me like a use case or a case study?
是的，比如我们在金融服务领域做了很多工作，因为你可以想象，这些数据在信息敏感性方面非常高。
Yeah, I mean, we do a lot of work, for example, in financial services, because as you can imagine, a lot of that data is quite sensitive in terms of information.
我们看到的具体应用场景之一是金融分析。
Very concrete use cases we're seeing is for financial analysis.
所以，有没有这样的人，他们的工作就是为各种客户提供建议，需要整合多样化的数据，比如哪些信息与这位特定客户相关，哪些信息与当前的市场环境、潜在可能性等相关？
So, have people whose job it is to advise various clients, and they need to pull on a diverse set of data, like what's the information that's relevant to this particular customer, what's the information that's relevant in terms of the current landscape, the possibilities, and so on?
将所有这些信息整合起来，为客户制定一份个人财务计划，正是这项技术能够大大简化的一种应用场景。
And kind of pull all of that information to make up a personal plan, a financial plan for a client is the kind of application that this technology can make much easier.
然后你可以查询你的计划，判断：我是否掌握了足够的信息？
And you can essentially then query your plan, decide, do I have enough information?
我是否还需要收集更多的信息来源？
Do I need to gather more sources of information?
你可以将内部信息与外部信息结合起来。
And you can combine the internal with the external information.
但这些信息的输出始终保持私密、安全，仅限于需要查看这些信息的人掌握。
But the output of that stays private, it stays secure, it stays in the hands of just the people who need to see that information.
我很高兴你提到这一点，因为最近有位金融行业的人问我，那些原本负责收集和整合外部信息的初级员工，未来会怎样？
And I'm glad you brought that up, because I was asked recently by someone in the financial service industry, what's going to happen to entry level employees, who were doing a lot of that, you know, collating and pulling in the external information.
我当时并没有一个很好的答案。
And I didn't have a great answer.
你知道，因为入门级员工的工资比普通员工低。
I I, you know, because, you know, you pay you pay entry level employees less than your standard employees.
你预期他们在工作中会有一些学习过程，并能完成一些有产出的任务。
You you anticipate there's gonna be some learning on the job and some productive things.
现在的问题是，如果这些员工……
And now the question is, what are these people going to do if they
如果这些入门级员工能够正确使用人工智能，他们就能直接跃升到能够真正胜任分析师的水平，借助工具完成十倍的工作量。
If can do it for these entry level employees are able to use AI properly, they're skipping ahead to the level where they can actually be fully functioning analysts and they can essentially do 10x the job with the tools.
因此，通过提供人工智能工具，他们在为雇主创造价值方面的能力得到了极大提升。
And so their growth in terms of their ability to deliver value to the employer has just been magnified by giving them the AI tools.
那么真正的威胁其实是针对中层员工吗？那些中年职场人士，他们可能会被取代，就像那个社交媒体实习生突然开始负责公司的公关或市场营销一样。
So is in the threat really to the middle, the people who are mid career who are going to get, I mean, it's like, it's the old story of the social media intern who comes in and all of a sudden is managing like PR or marketing for a company.
会不会是那些懂如何提示、会使用Cohere等工具的Z世代年轻人，让那些在某个岗位上干了十五年的人不得不回头张望？
Is it the Gen Z kid who like uses, who knows how to prompt and can use cohere and all of a sudden the person who's been doing things for fifteen years in a certain way has to look over their shoulder?
我认为，每当引入一种完全颠覆性的技术时，这种情况就会大量出现。
I do think that whenever you introduce a completely disruptive technology, that is a lot of what you see.
你会发现年轻一代对这种技术天生熟悉且直觉敏锐，他们真的能很快学会如何使用它。
You see the younger generation for whom that technology is native and is very intuitive and they really, you know, learn how to use it very quickly.
这让他们变得高效且富有成效得多。
And that just makes them so much more effective and productive.
而那些无法迅速掌握这项技术的人则发现自己处于劣势。
And folks who are not able to engage with a technology as quickly are finding themselves at a disadvantage.
我只是想起自己职业生涯早期，也许这就是为什么我没在公司待太久而不得不自己创业的原因，但那时充满干劲，渴望做事。
I just remember being early in my career and maybe this is why I didn't last very long in a company and had to go start my own, but having the energy and wanting to do things.
如果我当时有能构建原型的东西，我可以把它带到会议上展示，而不是请求开发人员花几个小时在这个副业上来改变局面？
And if I would have had something that could build a prototype and I could bring that to the meeting and show it as opposed to like, can I have like a couple hours of the developers time to work on this side project that would change things?
确实如此。
Absolutely.
而且说实话，这种能力对公司里的任何人都是开放的，对吧？
And to be honest, right, like that capability is afforded to anyone in the company, right?
不仅仅是初级员工才有权限使用它。
It's not just the more junior staffers that have access to it.
那些身居领导岗位的人，也不再需要写备忘录，而是可以直接制作出完整的原型。
It's also the people who are in leadership position, instead of like writing out a memo suddenly can go out and like produce a full fledged prototype.
他们不需要十个人、十名员工来帮他们制作原型。
They don't need, 10 people, 10 staffers to help them produce their prototype.
他们有了想法，就能快速制作出原型，然后发送给团队，推动项目启动。
They have an idea, they can quickly prototype it, and they send that to the team to get moving with the project.
我认为这种能力将为整个组织内设立项目开辟新的方式。
I think that kind of capability is going to open up new ways to set up projects across the organization.
这个云代码功能一直很值得观察。
This Cloud Code thing has been interesting to watch.
它一夜之间从一个能自动补全开发者代码的工具，变成了能上网执行任务、构建东西来完成特定目标的工具。
It went overnight from something that will auto complete developers code to like, we'll go out on the internet and do things and build things to accomplish specific tasks.
所以，AI系统外出执行任务这个想法，一方面，我听到过这样的故事，我也在节目中提过几次，比如前亚马逊全球消费者业务CEO在周末用振动编码搭建了一个CRM系统。
So is this idea of AI systems going out and doing things like on one hand, you know, I hear I see the story of, like and I've said this on the show a couple of times, but, you know, the former Amazon CEO of worldwide consumers going out and vibe coding a CRM over the weekend.
是的。
Mhmm.
你知道吗，这真的很酷。
You know, that's that's cool.
但我也在想，这到底有多真实呢？
But I'm also just like, you know, how real is that?
所以我很好奇。
So I'm curious.
哦，明白了。
Oh, okay.
你看着我的眼神好像在说：是的，这是真的。
You're giving me a look like, yes, it is real.
嗯，我认为这回到了我的观点，对吧？
Well, I think that goes back to my idea, right?
比如，能够以这种方式快速原型的人，并不意味着你周末随便写出来的代码会立刻变成一亿美元的生意，对吧？
Like those who are able to prototype in this way, it doesn't mean that whatever you've vibe coded into a weekend suddenly turns into a $100,000,000 business, right?
但它是一种向团队传达你意图的方式。
But it's a way to communicate with your teams, your intention.
所以只要你有好点子，就能以更真实的方式分享这些想法，并更快地开始原型设计。
So as long as you have good ideas, you're able to share these ideas in a way that much more real and to start prototyping much faster.
当然，还有其他方式来传达你的想法、指导团队，但这种方式突然打开了更多可能性。
Now, there's other ways to communicate your ideas, there's other ways to direct your teams, but that suddenly opens up so much more.
有趣的是，人工智能其实包含很多方面，但它本质上是一种沟通技术。
It is interesting how AI is AI is many things, but it's a communication technology.
它正逐渐变成这样。
It's becoming that right.
这正是新的编码代理正在开启的这类可能性。
And this is the kind of thing that that the new the new coding agents are opening up.
Cohere有没有在开发类似的东西？
Is Cohere does Cohere have like a version of this that it's working on?
我会说，是的，我们正在开发类似的功能。
I would say like, yes, we're working on the same kind of capabilities.
我们正在构建核心的通用模型，我认为目前我们在North平台提供的体验还是有些不同。
We're building core generic models, I would say that's, it's a bit of a different experience right now, that we're offering in terms of the north platform.
但这种协作工作还有很多。
But there is a lot of that sort of collaborative work.
有很多这样的情况，比如外出部署代理、利用外部信息。
There's a lot of this, you know, going out in deploying agents, leveraging external information.
所以有一些相似的元素，但目前我们对编码用例的关注度较低。
So there's some elements that are similar, but we're less focused specifically on coding use cases for now.
Cohere 显然已经筹集了大量资金，超过十亿美元。
Cohere obviously has raised a lot of money, more than a billion dollars.
但我只是想把这一点说清楚。
But I I'm this is not I'll just like draw it out.
OpenAI 一个周末的动静就这么大。
OpenAI sneezes that over a weekend.
现在这个世界里，人工智能是由少数几家超大型公司开发的。
You have a world now where AI is being developed by a handful of very big companies.
你之前的雇主 Meta 也是一个重要玩家。
Your former employer, Meta, is a big player.
亚马逊、谷歌、微软，当然还有OpenAI和Anthropic这些公司。
Amazon, Google, of course, Microsoft, and then OpenAI and Anthropic with these.
它们现在一年内就筹集了全部的风险投资资金。
They raise the entire year's worth of VC money in around now.
你如何看待这种高度集中于少数几家公司手中的风险？
What do you think about the risk of the fact that so much of this is being concentrated in so few hands?
老实说，我认为让多个团队能够开发和部署模型对整个生态系统是有益的。
Honestly, I do think it's beneficial to the ecosystem to have multiple groups who are able to develop models and to deploy them.
举个具体的例子，Cohere很早就开始研究多语言模型。
Think, you know, just to give you a concrete example, right, Cohere was very early on working on multilingual models.
也就是说，能够理解、处理多种语言——二十种、三十种甚至更多语言的信息的能力。
So, the ability to understand information, digest information across multiple languages, twenty, thirty, and so on languages.
我们有一系列备受推崇的AIA模型，而且是开源的等等。
We had a line of AIA models that is really well respected, open sourced and so on.
但对于那些极度专注于英语信息的公司来说，这根本不在他们的关注范围内。
It's just not on the radar of some of these companies that are very focused on English centric information.
完全没问题，你知道，不同公司有不同的领域，当我们进入亚洲市场、欧洲市场时，拥有一个在多种语言或本地语言上真正达到前沿水平的模型就变得至关重要。
Completely fine, you know, different space for different companies, when we get into markets in Asia, when we get into markets in Europe, suddenly, it matters to have a model that is actually state of the art across languages or across the local language.
因此，这现在开辟了完全新的市场。
And so that opens up completely new market right now.
机会如此广泛，以至于新兴玩家实际上仍有空间持续成长，获得健康的收入，吸引人才，并构建出与这些大公司不同的新东西。
Opportunities are so broad that actually there's space for, you know, up and coming players to really keep on growing, to have a very healthy revenue, to bring in talent, to actually build new things that are different from some of these other companies are building.
因此，我倾向于认为，有更多公司而非更少公司参与AI建设是极其健康的。
So, I tend to think it's super healthy to have more rather than fewer companies that are building AI.
我认为我们正看到这样的事实：回到我之前的观点，即许多不同的AI做着不同的事情，即使在公司层面，这种情况也在发生。
And I think we're seeing the fact that, you know, going back to my idea of, you know, many different AIs who do many different things, even at the company level, this is what's happening.
有许多参与者正在构建不同的东西，并相互学习。
A number of players who are building different things and learning from each other.
但大科技公司掌握了这么多，你不担心吗？
But the fact that big tech has so much of it, not a worry?
这并不让我担心。
It doesn't worry me.
好的。
Okay.
我的意思是，关于这个话题我们本可以聊得更久，但 Cohere 所构建的东西有着极其出色的成功路径，这完全不会让我失眠。
And I mean, you know, could have a much longer discussion about it, but it doesn't cause me to lose any sleep over the fact that like what we're building at Cohere has like an amazing, amazing path to be successful.
好的。
Okay.
顺便说一下，我提到了 Anthropic 和 OpenAI，而微软、亚马逊和谷歌都持有巨额股份。
By the way, I mentioned Anthropic and OpenAI, have Microsoft and Amazon and Google have massive stakes.
还有更多公司呢。
And there are many more.
是的。
Yes.
有个人叫 Dario Amodei，来自 Anthropic。
Somebody that does Wario, Dario Amade from Anthropic.
也许他从谷歌和亚马逊拿到那么多十亿资金这件事，并不是重点。
Well, maybe not the fact that he got all those billions from Google and Amazon.
但他确实对这些科技巨头有一些看法。
But he does have some things to say about the big tech companies.
这是他最近说过的一句话。
Here's a thing that he said recently.
这些公司中的一些是由具有科学背景的人领导的。
Some of these companies are essentially led by people who have a scientific background.
这正是我的背景。
That's my background.
Stemis Sabbas 的背景是谷歌DeepMind。
Stemis Sabbas's background Google DeepMind.
还有一些公司是由从事社交媒体的那一代企业家领导的。
Some of them are led by the generation of entrepreneurs that did social media.
科学家们长期以来都倾向于思考他们所创造技术的影响，并不回避责任。
There's a long tradition of scientists thinking about the effects of the technology they built not ducking responsibility.
我认为企业家的动机，尤其是社交媒体那一代企业家的动机，非常不同。
I think the motivation of entrepreneurs, particularly the generation of the social media entrepreneurs are very different.
他们与消费者互动的方式，可以说是一种完全不同的操控手段。
The way they interacted, you could say manipulated consumers is very different.
所以基本上，别以为他希望他们如此。
So basically, don't think he wants them
达里奥发表了强烈的观点。
Strong opinions from Dario.
人工智能，这对我来说，对达里奥·阿马德来说并不出乎意料。
AI, which is I guess not something out of character for Dario Amade.
你觉得这是一个合理的担忧吗？
Do you think that's a legitimate concern?
因为这太有趣了。
Because it's so interesting.
你是一位研究科学家，也曾就职于一家社交媒体公司。
You're a research scientist who also worked at a social media company.
所以如果有人知道答案，那一定就是你。
So if anyone knows the answer to this, it will be you.
我的意思是，我认为真正重要的是，没有人能在所有事情上都擅长。
I mean, I think what's really important, like no one is going to be good at everything.
对吧？
Right?
关键在于，你是否能让房间里其他人给你建议，帮助你打造出色的产品？
The question is like, do you get others in the room to advise you on how to build something great?
而且，你知道，我在Meta待过一段时间。
And, you know, I spent some time at Meta.
我会说，那里研究人员和领导团队之间有非常畅通的沟通渠道，意见都会被带到会议上。
I would say there very was strong channel from researchers to the leadership team and the opinions were brought into the room.
我认为，在Cohere我也确实看到过这种情况，研究团队、建模团队、产品团队，都有一个可以让这些不同观点汇聚在一起的场合。
I think, you know, I've seen that certainly at Cohere where the research team, the modeling team, product team, like there's a room where all these points of views can come together.
我回到这个想法：不能指望一个人掌握所有信息。
I go back to this thought, can't expect one person to all have all of that information.
只要他们组建了多元化的团队，并倾听这些不同的声音，最终就能打造出更好的产品。
And as long as they're building up the teams that are diverse, are listening to these diverse voices, like they will build better products at the end of the day.
好的。
Okay.
说到这里，随着广告开始进入生成式AI领域，像我这样的外行不禁好奇，这些公司是否会采取类似最大化用户参与度的策略，试图优化使用时长，从而提升相关数据指标。
On that note, as ads have started to enter the picture for generative AI, there's a wonder among outsiders like me about whether these companies will, do things like engagement max and try to optimize for time spent so they can, you know, get those numbers up.
我不想知道你是否认为这种情况会发生，但我想以研究者的身份问问你，这在经济上是否可行。
I don't want to ask you whether you think that's going to happen or not, but I want to ask you as a researcher, whether that's even economically feasible.
现在的模型是否已经足够高效，以至于比如你展示一条广告，用大语言模型来响应一次访问，也能成为一项盈利业务？
Are the models now efficient enough where like a let's say you were to show an ad, like a visit to to like to serve that visit with an LLM, could be a profitable thing.
还是说，提供这些服务的成本仍然如此高昂，以至于‘最大化参与度’这个概念根本站不住脚，因为从经济角度看它并不成立？
Or is it still so expensive to serve these use cases, that this even notion of engagement maxing doesn't make sense, because economically, it's not valid?
我的意思是，总体而言，通过不断试错，我们最终能找到可行的商业模式，对吧？
I mean, in general, right, like through trial and error, we find economic models that are viable, right?
这仍然是目前的现实。
Like that's still how it is.
这在很大程度上取决于定价模式等等因素。
It depends a lot on the pricing model and so on and so forth.
所以
So
而且广告可能很贵。
And might expensive ads to buy.
但你知道，这取决于模型的设置方式。
But you know, it depends, you know, it depends on how the model is set up.
所以我不确定这是否会是最初推出的方式，我们需要看看后续会如何发展。
So I don't know that this is the way that that gets rolled out initially, we'll have to see what's the progression of that.
你觉得，我们有能力根据已有信息定制内容，这从经济角度来看将是一个持续被利用的手段。
Do think, you know, we have the ability to tailor content based on the information we have that is there that is a lever that's going to continue to be used from an economic point of view.
在我们结束之前，谈谈AI主权吧，各国和银行等机构正开始构建自己的模型，而不是依赖现成的产品。
AI sovereignty, before we go, countries are starting and institutions like banks are starting to build their own models or they're not relying on off the shelf stuff.
所以谈一谈吧，因为这是Cohere正在做的事情。
So talk a little bit because this is something Cohere is working on.
这是我不太了解的事情，但确实存在这种趋势，或者至少这是一个正在被讨论的话题。
It's something I don't know a lot about the fact that there is this push or at least it's something that is being discussed.
那么，什么是人工智能主权，它如何体现？
So what is AI sovereignty and how is it playing out?
是的，主权这个词已经被用于几种不同的含义。
Yeah, sovereignty has been used in a few different ways.
在某些情况下，它意味着拥有自己的模型。
In some cases, it means the ability to have your own model.
因此，在金融服务和银行领域，这确实是它们投入大量时间、思考和寻找解决方案的领域。
So in the case of financial services and banks, that is definitely something that they spend a lot of time investing in, thinking about, looking for solutions.
它们看到了其中的机会。
They see the opportunity.
我认为，它们甚至是前几代人工智能技术的早期采用者，比如预测模型、统计模型等等。
They were, I think, early adopters even of previous generation AI technologies, predictive models, for example, statistical models, and so on.
因此，它们将这视为自然的演进。
And so they see this as the natural evolution.
因此，它们在人工智能的复杂性和准备程度方面已经相当先进。
So, they're pretty advanced in terms of their sophistication and their readiness for AI.
而且他们通常有资金投入其中。
And often they have the means to invest in it.
因此，我们确实看到那里有很多兴趣。
And so, we're definitely seeing a lot of interest there.
但通常来说，我认为人才缺口让他们面临一些困难。
Often though, you know, I think the talent gap makes it a little bit harder for them.
所以有时他们尝试自己构建模型，然后来找我们，希望找到一些更成熟、开箱即用的解决方案。
So sometimes they've tried to build their own models and so on, and they come to us and they're looking for solutions that are a little bit more mature out of the box and so on.
因此，我们在那里建立了非常稳固的合作伙伴关系。
And so we have really solid partnerships going on there.
我们听到的另一种关于主权的解读是，企业希望制定一个稳健的AI战略。
The other way to think about sovereignty that we're hearing a lot is that companies want a robust plan for AI.
所以，他们希望拥有多种选择。
And so, you know, they want options.
他们可能正在使用一个模型，但实际上希望拥有另一个模型来进行比较和基准测试。
They may be using one model, but they actually want to have another model to be able to compare, to benchmark.
如果一种模型的访问被切断或过于昂贵，他们还有另一种选择。
If one model access gets cut off or too expensive, they have another one.
因此，主权的一个方面实际上是关于制定一个稳健的战略。
And so, there's an aspect of sovereignty that's really about building a robust strategy.
这不仅仅是使用自己的或单一的工具，而是关于对技术访问的控制。
It's not about just using your own or using one thing, but it's about having control over the access to the technology.
是的，正如你所说，对我来说，这一切进展得实在太快了。
Yeah, it's just as you speak about it, to me, it's just amazing how fast this has moved.
回想我们2022年的第一次会面，现在已经是2026年了。
Going back to our first meeting in 2022, the fact that we're it's 2026.
所以已经过去了三年多。
So it's been three years and change.
但每年之间的变化简直天差地别。
But it's it's just a world of a difference year to year to year.
对。
Yeah.
所以最后一个问题给你，这种速度能保持下去吗？
So last question to you, can the pace keep up?
在许多方面，进展仍然非常迅速，比如投资的规模。
It is still moving very fast on so many fronts, you know, just the size of the investments.
我认为在采用方面，我们还处于曲线的早期阶段。
I think on adoption, we are so early in the curve.
因此，下一个挑战是如何让这项技术成功地渗透到社会、商业世界和人们的生活中。
And so that's going to be the next challenge to see how do we enable this technology to sort of disperse through society, through the business world in people's lives, how do we do that successfully.
但是的，我认为在商业化和采用方面，目前还处于非常非常早期的阶段。
But yeah, I think the pace, especially when it comes to commercialization and adoption, is really very, very early days.
还有很长的路要走。
Got a long way to go.
真的吗。
Seriously.
好了，Joelle，我们已经聊过好几次了。
Well, Joelle, we've spoken a handful of times.
我总是很欣赏你能将我们许多人所关心的宏大议题，扎根于研究和实际应用之中。
I always appreciate how you're able to take these big things that a lot of us are wondering about and grounded in the research and the practical side of things.
所以你随时欢迎再来节目，感谢你今天莅临。
So you're always welcome on the show and thank you for coming on.
谢谢。
Thank you.
每次都非常愉快。
Always a pleasure.
好了，各位。
Alright, everybody.
非常感谢大家的观看和收听，也感谢高通公司让我们在达沃斯的这个场地录制，我们下次再见于《科技大观》播客。
Thank you so much for watching and listening and thank you to Qualcomm for having us here at the space at Davos and we'll see you next time on big technology podcast.
好了。
Alright.
谢谢。
Thank you.
太棒了。
That was great.
非常感谢。
Thank you so much.
谢谢。
Thank you.
关于 Bayt 播客
Bayt 提供中文+原文双语音频和字幕，帮助你打破语言障碍，轻松听懂全球优质播客。