1
00:00:00,120 --> 00:00:03,560
So I hope that you are gobsmacked by this.
所以我希望你对此感到惊讶。

2
00:00:03,800 --> 00:00:05,560
This was really, really cool.
这真的非常非常酷。

3
00:00:05,800 --> 00:00:08,120
Uh, I think I just just counted it up.
呃，我想我只是算了一下。

4
00:00:08,120 --> 00:00:16,480
I think that in total running this framework involved, I've counted this upright, uh, 34 model calls
我认为在运行这个框架的过程中，我统计了 34 个模型调用

5
00:00:16,480 --> 00:00:18,040
in total to do all of this.
总共完成所有这一切。

6
00:00:18,080 --> 00:00:26,760
29 Lrms, five, uh, neural network calls, um, including each of the parts of that agent loop as
29 Lrms，五个，呃，神经网络调用，嗯，包括该代理循环的每个部分

7
00:00:26,760 --> 00:00:27,720
a separate call.
单独的呼叫。

8
00:00:27,920 --> 00:00:34,000
Um, because it was like a different, different call to to to to um, uh, out of that, it's we were
嗯，因为这就像一个不同的、不同的呼叫 to to to 嗯，呃，从中，我们是

9
00:00:34,000 --> 00:00:39,560
calling three different frontier models GPT five, GPT five one, and anthropic, uh, Claude, uh,
称三种不同的前沿模型为GPT 5，GPT 5 1，以及人择，呃，克劳德，呃，

10
00:00:39,560 --> 00:00:40,280
45.
45.

11
00:00:40,600 --> 00:00:46,280
And we were calling three open source models, our fine tuned specialist model, uh, the all mini LM,
我们调用了三个开源模型，我们微调的专业模型，呃，全迷你 LM，

12
00:00:46,280 --> 00:00:55,480
l6, v2, uh, encoder from Hugging Face and OSS, uh, that we used for GPT oss 20 that we use for
l6，v2，呃，来自 Hugging Face 和 OSS 的编码器，呃，我们用于 GPT oss 20

13
00:00:55,480 --> 00:00:56,520
pre-processing.
预处理。

14
00:00:56,520 --> 00:00:59,490
So three frontier, three open source models.
所以三个前沿，三个开源模型。

15
00:00:59,610 --> 00:01:00,690
29.
29.

16
00:01:00,730 --> 00:01:08,610
LMS plus five neural network calls to to our deep neural network for a total of 34 model calls, all
LMS 加上五个神经网络调用我们的深度神经网络，总共 34 个模型调用，全部

17
00:01:08,610 --> 00:01:16,970
collaborating together in order to achieve this result of finding a Dell 16 Plus Ultra that got sent
共同协作以实现找到已发送的 Dell 16 Plus Ultra 的结果

18
00:01:16,970 --> 00:01:21,650
to me as a push notification that that is quite a platform.
对我来说，这是一个推送通知，这是一个相当不错的平台。

19
00:01:21,930 --> 00:01:23,250
It's quite a product.
这是一个非常好的产品。

20
00:01:23,370 --> 00:01:25,210
Uh, so I hope you enjoyed this.
呃，所以我希望你喜欢这个。

21
00:01:25,210 --> 00:01:27,490
I hope you've been able to get this to work yourself.
我希望您自己能够做到这一点。

22
00:01:27,490 --> 00:01:32,650
Remember, if you hit any technical snafus because there's a lot going on here, you can comment stuff
请记住，如果您遇到任何技术问题，因为这里发生了很多事情，您可以发表评论

23
00:01:32,650 --> 00:01:32,890
out.
出去。

24
00:01:32,890 --> 00:01:36,570
You can reduce the complexity of the ensemble model.
您可以降低集成模型的复杂性。

25
00:01:36,570 --> 00:01:39,530
Just just calling one of those models for sure.
只需调用其中一个模型即可。

26
00:01:39,810 --> 00:01:44,490
Uh, you can simplify down in many different places if all of the moving parts aren't working for you,
呃，如果所有的活动部件都不适合你，你可以在很多不同的地方进行简化，

27
00:01:44,490 --> 00:01:45,450
but I hope they are.
但我希望他们是。

28
00:01:45,530 --> 00:01:47,370
And I hope you're seeing something like this.
我希望你看到这样的事情。

29
00:01:47,570 --> 00:01:52,610
And if you just want, if you're just in it to get the intuition, then it's just important to see that
如果你只是想要，如果你只是为了获得直觉，那么看到这一点就很重要

30
00:01:52,610 --> 00:01:59,050
it really wasn't that hard to build this sophisticated platform with six different 6 or 7 different
用六种不同的 6 种或 7 种不同的方法构建这个复杂的平台确实并不难

31
00:01:59,050 --> 00:02:02,770
models collaborating on this across 34 different calls.
模型在 34 个不同的调用中对此进行协作。

32
00:02:02,890 --> 00:02:05,250
Uh, with this fantastic outcome.
呃，有了这个奇妙的结果。

33
00:02:05,290 --> 00:02:05,850
All right.
好的。

34
00:02:05,850 --> 00:02:11,730
I will see you for the wrap and take one more look at the architecture that we built and, uh, take
我会在总结中见到您，并再看一下我们构建的架构，然后，呃，采取

35
00:02:11,730 --> 00:02:13,010
great pride in it.
对此深感自豪。

36
00:02:13,010 --> 00:02:16,850
And if you've managed to tweak it and do anything different, I would love to hear about it.
如果您设法调整它并做任何不同的事情，我很想听听。

37
00:02:16,850 --> 00:02:19,210
This is, uh, been so much fun building this.
呃，构建这个非常有趣。

38
00:02:19,210 --> 00:02:22,010
And you're probably thinking, was that that was the finale, right?
你可能会想，这就是结局吧？

39
00:02:22,010 --> 00:02:23,730
That was that was really great.
那真是太棒了。

40
00:02:23,770 --> 00:02:25,210
No, there's still more.
不，还有更多。

41
00:02:25,250 --> 00:02:26,530
There's still a little bit more.
还有一点点。

42
00:02:26,530 --> 00:02:28,890
We've got one day left of this thing.
这件事我们还剩一天了。

43
00:02:28,890 --> 00:02:29,810
One more day.
还有一天。

44
00:02:29,970 --> 00:02:35,810
It's a short day, but it's a great day and it will be our final, final finale, the finale of finales.
这是短暂的一天，但却是伟大的一天，这将是我们的最后、最后的结局、结局中的结局。

45
00:02:35,850 --> 00:02:42,650
We will finish the Agentic workflow by adding in things like memory and stuff like that, and by the
我们将通过添加诸如内存之类的东西来完成 Agentic 工作流程，并且通过

46
00:02:42,650 --> 00:02:47,970
end of it, not only will you be able to do everything on this slide that I'm not going to say yet again,
结束后，您不仅能够完成这张幻灯片上我不会再说一遍的所有操作，

47
00:02:48,010 --> 00:02:52,730
you'll be able to do all of this, and you'll also be able to say, you know what?
你将能够做到所有这些，而且你还可以说，你知道吗？

48
00:02:52,770 --> 00:02:56,250
At this point, I am an AI engineer.
此时，我是一名人工智能工程师。