1
00:00:04,840 --> 00:00:13,900
In this lesson, we are going to talk about LMS and about three things around LMS that are important

2
00:00:13,900 --> 00:00:19,000
the size, the precision and the cost of the LM.

3
00:00:25,610 --> 00:00:34,040
These three things are important about LMS, especially at this point, because some of you may have

4
00:00:34,040 --> 00:00:42,650
heard that the thing to do at this point is to create your own LM, and this is not accurate.

5
00:00:42,740 --> 00:00:51,980
A and just think about the size of what we call the foundational LMS, which are what we call LMS in

6
00:00:51,980 --> 00:00:53,000
this program.

7
00:00:53,780 --> 00:01:01,580
So a typical person reads between 0 and 700 books in their lifetime.

8
00:01:02,930 --> 00:01:08,450
A book has eight 80,000 words.

9
00:01:10,070 --> 00:01:18,080
So ChatGPT has been trained with the text equivalent to 10 million books.

10
00:01:18,770 --> 00:01:23,990
That is 800,000 million words.

11
00:01:26,020 --> 00:01:35,500
Remember, LMS have been made possible thanks to deep learning, neural networks, transformers and

12
00:01:35,500 --> 00:01:36,520
GPUs.

13
00:01:37,910 --> 00:01:48,590
So just think about the computational power you need to train ChatGPT for.

14
00:01:50,220 --> 00:01:52,200
With the amount of data.

15
00:01:52,820 --> 00:01:54,470
We are talking about.

16
00:01:55,600 --> 00:01:57,520
Like 10 million books.

17
00:01:59,640 --> 00:02:03,660
Let's think about the precision of LMS.

18
00:02:05,510 --> 00:02:14,390
The accuracy of LMS depends on the quality of the data it has been trained on and the amount of data

19
00:02:14,390 --> 00:02:16,010
it has been trained on.

20
00:02:17,090 --> 00:02:18,080
So.

21
00:02:19,260 --> 00:02:24,840
It depends on the quality and quantity of the data it has been trained on.

22
00:02:26,460 --> 00:02:33,600
The position is nowadays the most important feature of any LM.

23
00:02:33,990 --> 00:02:41,910
And that's why during 2023, the most used LM was ChatGPT.

24
00:02:42,990 --> 00:02:44,760
We will talk more about that.

25
00:02:44,790 --> 00:02:49,770
Remember, LM applications are not LMS.

26
00:02:50,460 --> 00:02:57,960
LM applications are applications we build on top of LMS like ChatGPT.

27
00:02:59,340 --> 00:03:09,510
In order to create ChatGPT or any other foundational LMS as they are called, you need huge tons of

28
00:03:09,660 --> 00:03:20,490
data and you need a very refined way of doing things in order to get the precision level that ChatGPT

29
00:03:20,520 --> 00:03:21,510
has got.

30
00:03:22,380 --> 00:03:34,200
And if you think it's not enough, then you may think about the cost of training ChatGPT for the CEO

31
00:03:34,200 --> 00:03:35,100
of OpenAI.

32
00:03:35,130 --> 00:03:45,330
OpenAI said recently that training ChatGPT for costed more than $100 million.

33
00:03:46,230 --> 00:03:47,250
So.

34
00:03:47,910 --> 00:03:54,300
We understand now that LMS because of its size.

35
00:03:55,790 --> 00:03:59,690
The precision and the cost they require.

36
00:04:01,070 --> 00:04:09,020
Excuse me, are very difficult things to achieve, to build.

37
00:04:09,020 --> 00:04:12,980
And this is not what you should do.

38
00:04:12,980 --> 00:04:16,070
Focus your attention on building an LM.

39
00:04:17,089 --> 00:04:23,000
We are going to build LM applications on top of LMS.

40
00:04:24,780 --> 00:04:32,130
In the next lesson we will understand the meaning of foundation LM models.