1
00:00:00,040 --> 00:00:02,360
So what are some of the typical ways people use tools?
那么人们使用工具的典型方式有哪些？

2
00:00:02,360 --> 00:00:06,360
Well, obviously they're very useful for looking things up in a database.
嗯，显然它们对于在数据库中查找内容非常有用。

3
00:00:06,400 --> 00:00:07,480
I mentioned that already.
我已经提到过这一点。

4
00:00:07,840 --> 00:00:12,480
They're useful for actually taking actions you could get in a conversation with NLM.
它们对于实际采取您可以在与 NLM 对话中获得的行动非常有用。

5
00:00:12,520 --> 00:00:16,040
An assistant could book a meeting, book airline tickets.
助理可以预订会议、预订机票。

6
00:00:16,440 --> 00:00:20,960
You could have your LLM be able to to carry out a calculation.
您可以让您的法学硕士能够进行计算。

7
00:00:21,000 --> 00:00:24,680
Something that llms are famously bad at is doing any kind of maths.
众所周知，法学院最不擅长的就是数学。

8
00:00:24,680 --> 00:00:30,520
Well, you can give it a tool that can do the maths for it, as we suspect ChatGPT is given tools like
好吧，你可以给它一个可以进行数学计算的工具，因为我们怀疑 ChatGPT 提供了类似的工具

9
00:00:30,520 --> 00:00:33,400
that, which is why it's now able to do calculations easily.
这就是为什么它现在能够轻松地进行计算。

10
00:00:33,760 --> 00:00:39,720
Similarly, you could give it a tool that can execute Python code, perhaps in a secure way, like in
同样，您可以给它一个可以执行 Python 代码的工具，也许以安全的方式，就像

11
00:00:39,720 --> 00:00:41,360
a in a Docker container or something.
Docker 容器或其他容器中的 a。

12
00:00:41,400 --> 00:00:46,240
That means that it's ring fenced, sandboxed, so that that kind of when you do that, it's sometimes
这意味着它是被围起来的、沙盒的，所以当你这样做时，有时会

13
00:00:46,240 --> 00:00:48,440
called as having a coder agent.
称为具有编码器代理。

14
00:00:48,560 --> 00:00:54,760
It's it's an LLM that has the ability to execute code in order to carry out its task.
它是一个法学硕士，有能力执行代码以完成其任务。

15
00:00:54,960 --> 00:00:58,920
A coder agent doesn't always doesn't necessarily mean an agent that writes code.
编码代理并不总是意味着编写代码的代理。

16
00:00:58,920 --> 00:01:05,120
It can really mean an agent, an LLM call that's able to execute code in carrying out its task.
它实际上可以指一个代理，一个能够执行代码来执行其任务的 LLM 调用。

17
00:01:05,520 --> 00:01:12,160
And then finally for for this list would be an example of where you can have an LLM that can directly
最后，对于这个列表，您可以拥有一个可以直接获得法学硕士的示例

18
00:01:12,160 --> 00:01:18,160
modify the UI, it can produce a new chart that gets visualized instantly to the user, so it can do
修改 UI，它可以生成一个新的图表，立即向用户可视化，因此它可以

19
00:01:18,160 --> 00:01:20,640
things that can immediately interact with the user.
可以立即与用户交互的东西。

20
00:01:20,680 --> 00:01:23,920
And of course, there are so many different ways you can use tools.
当然，使用工具的方式有很多种。

21
00:01:23,920 --> 00:01:27,480
You can probably think of 100 others that I haven't put here because there are many, many, many.
你可能还能想到另外 100 个，但我没有在这里列出，因为还有很多很多。

22
00:01:27,760 --> 00:01:33,240
But importantly, there's a whole category in particular that I haven't mentioned yet.
但重要的是，有一个具体的类别我还没有提到。

23
00:01:33,360 --> 00:01:36,200
Uh, which is where the greatest excitement is.
嗯，这就是最让人兴奋的地方。

24
00:01:36,440 --> 00:01:39,480
And of course, it's all about a gigantic AI.
当然，这一切都与巨大的人工智能有关。

25
00:01:40,040 --> 00:01:47,040
There are, uh, two ideas, two ways you can use tools that form some of the the core ideas behind
有，呃，有两种想法，两种方法可以使用工具来形成背后的一些核心想法

26
00:01:47,080 --> 00:01:48,000
Agentic AI.
代理人工智能。

27
00:01:48,000 --> 00:01:54,840
And first of all, put simply, you could use a tool which itself involves another call to an LLM.
首先，简单地说，您可以使用一个本身涉及对法学硕士的另一次调用的工具。

28
00:01:55,120 --> 00:01:58,000
So you could say to an LLM, hey, you have three different tools.
所以你可以对法学硕士说，嘿，你有三种不同的工具。

29
00:01:58,000 --> 00:02:03,920
They do A, B and C, and actually A, B and C are all calls to different llms to take care of that
他们执行 A、B 和 C，实际上 A、B 和 C 都是对不同 llms 的调用来处理该问题

30
00:02:03,920 --> 00:02:04,600
task.
任务。

31
00:02:04,800 --> 00:02:11,920
And as a result, you're equipping an LLM with the ability to orchestrate the activities of other LLM
因此，您为法学硕士配备了协调其他法学硕士活动的能力

32
00:02:11,920 --> 00:02:12,560
calls.
来电。

33
00:02:12,560 --> 00:02:15,640
And that, of course, is one of the core ideas behind Agentic AI.
当然，这是 Agentic AI 背后的核心理念之一。

34
00:02:15,840 --> 00:02:22,520
It allows you to have agentic workflows where an LLM controls the The workflow and where it orchestrates
它允许您拥有代理工作流程，其中法学硕士控制工作流程及其编排

35
00:02:22,560 --> 00:02:23,560
activities.
活动。

36
00:02:23,560 --> 00:02:24,800
So that's one core idea.
这是一个核心理念。

37
00:02:24,800 --> 00:02:30,920
And then the other core idea is just this idea that you can have a tool which is like a to do list kind
然后另一个核心想法就是你可以拥有一个类似于待办事项列表的工具

38
00:02:30,920 --> 00:02:37,000
of tool, which allows an Lem to plan out a series of activities and then like, check them off and
工具，它允许 Lem 计划一系列活动，然后检查它们并

39
00:02:37,000 --> 00:02:42,480
then continually refine them and, and have some evaluation criteria that it needs to get to before
然后不断地完善它们，并且有一些之前需要达到的评估标准

40
00:02:42,600 --> 00:02:43,480
it's sure that it's done.
确定已经完成了。

41
00:02:43,480 --> 00:02:50,680
The list and all of that kind of functionality allows you to build an Lem system, which operates in
该列表和所有此类功能允许您构建一个 Lem 系统，该系统在

42
00:02:50,680 --> 00:02:55,840
a loop, able to take on activities and see them through to completion.
一个循环，能够进行活动并完成它们。

43
00:02:56,000 --> 00:03:01,840
And that whole idea of an agentic loop, of course, that is a core idea behind Agentic AI, and it's
当然，代理循环的整个想法是代理人工智能背后的核心思想，它是

44
00:03:01,840 --> 00:03:04,120
enabled by the use of tools.
通过使用工具来启用。

45
00:03:04,280 --> 00:03:08,040
And you saw it in a very tangible way with Claude code.
您可以通过克劳德代码以非常具体的方式看到它。

46
00:03:08,280 --> 00:03:13,440
So these, these ideas and many others, uh, all form some of the basis of Agentic AI.
所以这些，这些想法和许多其他想法，呃，都构成了智能人工智能的一些基础。

47
00:03:13,440 --> 00:03:18,440
And so you can see how essential tools are to, to agents.
由此可见，工具对于代理商来说是多么重要。

48
00:03:18,640 --> 00:03:21,520
And that is some of the common use cases of tools.
这就是工具的一些常见用例。

49
00:03:21,520 --> 00:03:22,640
But there are many more.
但还有更多。

50
00:03:22,800 --> 00:03:27,200
Uh, and I think that it's about time that we stopped all this talking stuff, of course.
呃，当然，我认为我们是时候停止所有这些谈话了。

51
00:03:27,360 --> 00:03:31,360
And we go to Cursa and we start working with tools.
我们去了 Cursa，开始使用工具。