Microsoft has been working on AI for decades and chatbots actually aren't anything new, but all of a sudden everyone is salivating.
微软已经在人工智能领域研究了几十年,聊天机器人其实也已经不是什么新鲜事了,但突然间每个人都对这一块战地垂涎三尺。
Why do you think the moment for AI is now?
你认为为什么有人会说现在就是人工智能发展的最佳时机?
Yeah, I mean it's actually you're absolutely right, which is AI has been here.
是的,实际上我觉得你是完全正确的,人工智能就在我们眼前。
In fact, it's mainstream, right?
事实上,这已经成为主流了,对吧?
I mean search is an AI product.
我的意思是搜索引擎其实也是一种人工智能产品。
Even the current generation of search, every news, aggregation, recommendation and you know YouTube or e-commerce or Tik Tok or all AI products, except they're all, I'll say, today's generation of AI is all autopilot.
即使是目前的搜索技术,每个新闻、聚合、推荐,包括油管、电子商务、抖音以及其他所有人工智能产品,但和这些不同的是,今天的人工智能是高度自驱的。
In fact, it's a black box that we just sort of use.
事实上,它就是我们使用的一个黑匣子。
That is dictating, in fact, how our attention is focused whereas going forward.
这决定了我们今后应将发展的重心置于何处。
The thing that's most exciting about this generation of AI is perhaps we move from autopilot to co-pilot, where we actually prompt it.
这一代人工智能最令人兴奋的地方可能是我们从主动操控转向了副操控的位置,所以我们实际上是在提示人工智能。
I mean, think about it right.
仔细想想我所说的。
What we we are learning to program AI as with just natural language.
我们正在学习用自然语言编程人工智能。
Right and it gets smarter every time you use it.
是的,而且每使用它一次,它就会变得更智能。
Yeah, and also you are making it.
是啊,可以说你是在在创造人工智能。
It's just it's ultimately a stochastic machine that you're using as a tool to help reason about what you're learning, what you're creating, what you're doing.
它只是一个随机机器,你把它作为一个工具来帮助你对所学的东西进行推理,帮助你创作,帮助你工作。
And so yes, I think this shift from autopilot to Copilot is actually yes the next phase of AI, which in fact, is perhaps going to put us as humans, you know, more in the center of using AI to our benefit.
所以,我认为从主动操控向副操控的转变实际上是人工智能的下一个阶段,事实上,通过这样的方式,作为人类来说,更重要的是利用人工智能为我们造福。
With Copilot you are deeply weaving LLMS across all of Microsoft's products, Word, Excel, PowerPoint, Outlook.
你可以借助Copilot将LLMS植入到微软的所有产品中,包括Word、Excel、PowerPoint、Outlook。
You're also basically giving folks their own personal business chatbot.
这基本上等于是给了人们一个专属于自己的个人商务聊天机器人。
How transformative a change do you think this will be in how we work?
你认为这会给我们的工作方式带来多大的变革?
Yeah, to me that is it.
是的,确实是这样。
Having built now GitHub Copilot having built, the Web Copilot with being and even what we did with dynamic.
现在GitHub Copilot已经建立了,Web Copilot正在建成中,我们在动态管理方面也做了很多的工作。
This is the big next step for us — to put it in the tools everybody uses every day for their work. I think it does three things, Emily.
这是我们迈出的重要一步——把它应用到每个人每天工作所使用的工具中。我觉得它有三个作用,艾米丽。
For me, you know, one of the things that I've always said is, God, there's so much functionality in Word or Excel and PowerPoint.
我经常会说:天哪,Word、Excel和PowerPoint有这么多功能。
How do we make it such that people use this in powerful ways to create great content, great documents, great PowerPoints art, learn how to do analysis.
我们如何让人们利用这些强大的工具来创造好的内容、好的文档、好的幻灯片艺术,以及学习如何做分析。
That's pretty sophisticated in an interesting way now without having to say let me learn all the commanding of Office.
这是一种相当复杂且有趣的方式,现在,我们不必去学习所有的Office操作指令。
I just literally can use natural language.
因为我们可以借助自然语言来实现。
So the power of 30 plus years of Office creation of the sophistication of these tools is just available to every user, same thing with even Teams and Teams Copilot.
因此,每个用户都可以使用Office 30多年来所创造的这些复杂工具的力量,同样的事情也发生在Teams和Teams Copilot上。
Like think about how meetings can be more effective with the Team Copilot.
比如,如何借助Team Copilot让会议更高效。
But I think the probably the biggest difference maker will be business chat.
但我认为最大的区别可能是商务洽谈。
Because if you think about the most important database in any company is the database underneath all of your productivity software except that data is all siled today.
因为,对于任何公司而言,最重要的数据实际就是生产力方面的数据,只不过现在所有的数据铺天盖地地向我们袭来。
But now I can query it with natural.
但现在,我可以用自然语言去询问。
I can say, oh, I'm going to meet this customer.
我可以说,我要去见这位顾客。
Can you tell me the last time I met them?
你能告诉我上一次会面是什么时候吗?
Can you bring up all the documents that are written up about this customer and summarize it, so that I'm current on what I need to be prepped for.
你能把有关这位客户的所有文件调出来并总结一下吗?这样我就能知道我需要准备些什么了。
That ability to interrogate that database query it and do it without learning some new syntax of querying language, but just natural language.
这种查询数据库的能力不需要我们学习查询的语言,只需要使用我们的自然语言即可。
It's just super powerful.
功能超乎强大。
How do you make sure it's not Clippy 2.0?
如何确保它不变成第二个Clippy?
That it is helpful ,delightful doesn't want to make me click out ASAP.
Clippy确实很有用,而且很有趣,让人点进去就不想马上退出。
There are two sets of things.
有两种情况。
One is you...... You're laughing because?
其中一方面......你笑什么?
Because look like our industry is full of lots of, you know, examples from Clippy to even like say current generation of these assistants and so on.
因为,我们的行业似乎并不乏像Clippy一类的产品。
They all are brittle.
但这些产品并没有那么强大。
I think there are two things that this generation of AI do.
我认为这一代人工智能能做两件事。
One is when we say "they understand natural language", they truly understand natural language.
一是当我们说人工智能可以理解自然语言,实际情况真的就是如此。
We are also going to have to learn that ultimately these are tools.
我们还必须明白,人工智能最终还是一种工具。
They're stochastic in nature, just like any time somebody sends me a draft.
本质上来讲,它们的输出是随机的,就好像我在看别人的创作草稿。
I review the draft.
我会对其进行研究。
I just don't accept the draft.
但我不会轻易就接受。
We will do that like interestingly enough.
每个人都会有不同的方式来面对人工智能给出的答案。
We learned a lot in GitHub Copilot.
我们从GitHub Copilot上学到了很多。
In fact, the day first time when get up Copilot came, you know, even software developers saying "oh yeah, this does make mistakes."
事实上,当我们第一次开始对人工智能进行提示时,即使是软件开发人员也会说:“哦,是的,这确实会出错。”
Except in even few months people said "oh yeah, but I know how to correct those mistakes."
但只要几个月的时间,人们就会说:“哦,人工智能确实会出错,但我知道如何去纠正这些错误。”
And so that ability to work with this Copilot give it feedback, know how to verify it, even this chain of thought reasoning in Excel like the one feature.
因此,借助Copilot所带来的反馈,我们就知道如何去纠正这些错误,我们甚至可以运用这样的方式来操作Excel。
I don't know if you saw this, but that was really cool, which is we said "okay, what's the design choice we can make so that users get in the habit of not just accepting whatever AI is saying.
我不知道你有没有注意到一个很有意思的地方,有时候,研发人员会思考:“我们应该怎样设计人工智能,从而让用户养成不只是单纯接受人工智能输出的习惯。”
But even ask it to show you its scratch pad work.
而是能够让它向你展示它的工作流程。
Right.
没错。
It's like literally like inspecting somebody's homework, which is hey tell me exactly how you did this and so that I can verify.
就像你在看别人作业的时候,你会问那个人:“告诉我你是怎么做的,让我来验证一下。”
Those are the kinds of things that we'll learn.
这些都是我们需要学习的
In my decade plus covering Microsoft, I can't remember you releasing this much in quick succession.
在我报道微软的十多年里,你发布的产品实在是多到我都记不过来。
Why is it all happening so fast?
为什么会这么快?
It's sort of sometimes it feels it's all happening fast.
有时候感觉一切都发生得太快了。
It's we started working on this a good four years ago.
实际上我们在四年前就开始做这件事了。
I mean in some sense if you think about when OpenAI and Microsoft came together and said, "hey, this next generation of large language models need new infrastructure.
在某种意义上,可能是因为OpenAI和微软都认为,我们的下一代需要一套更新的大型语言模型设施。
Let's build the infrastructure, tune the infrastructure.
让我们把它建立起来,再去调整它。
Let's understand even what AI safety and alignment looks like for these.
让我们来了解一下人工智能的安全性和一致性究竟能做到什么样子。
What are the user's cases.
用户的体验如何?
And this has been four years plus in making.
这些工作花了我们四年多的时间。
So once we started seeing the scaling effects the promise of the emergent capabilities, even that started showing up in these large language models.
因此,一旦我们开始看到规模效应和新兴技术带来的希望,我们便看到这些技术所带来的改变在大型语言模型中也开始显现出来。
That's why last year, in fact, that perhaps for me, application of these large language models inside or GitHub Copilot was the big... it's the biggest LLMS product out there today.
这就是为什么直到去年,对我来说,在这些大型语言模型应用程序中,GitHub Copilot的体量真的大到…可以说是目前最大的LLMS产品。
And so that gave us confidence.
这给了我们信心。
We now can apply it in more context.
现在我们可以把它应用到更多的场景中。
So, yes, it feels that we're launched a lot of things just in a hurry this year.
所以,没错,感觉我们今年推出了很多产品。
But it's been four years in the making and obviously it's a great partnership with OpenAI.
但实际上,这已经酝酿了四年之久,显然这是与OpenAI的一次伟大合作。
还没有评论,快来发表第一个评论!