1
00:00:05,340 --> 00:00:16,440
In this lesson, we are going to talk about the selection of a foundation LM when we are a building,

2
00:00:16,440 --> 00:00:17,610
an LM application.

3
00:00:17,610 --> 00:00:25,410
So one of the first thing, first thing we need to decide is which foundation LM are we going to use.

4
00:00:25,410 --> 00:00:32,910
Are we going to use ChatGPT or or llama two or anthropic or Mistral.

5
00:00:32,910 --> 00:00:41,640
What is the the what are the criteria we are going to use in order to select the the foundation LM.

6
00:00:45,870 --> 00:00:48,210
So in the.

7
00:00:49,140 --> 00:00:51,060
2023.

8
00:00:52,210 --> 00:00:56,200
Artificial Intelligence Engineers survey.

9
00:00:57,040 --> 00:01:06,100
One of the questions we found there was the top considerations when choosing a foundation model for

10
00:01:06,100 --> 00:01:14,500
your LM application, and the four top considerations when choosing foundation models were.

11
00:01:15,430 --> 00:01:16,150
Accuracy.

12
00:01:16,180 --> 00:01:19,180
The first one cost.

13
00:01:20,120 --> 00:01:22,880
Latency and privacy.

14
00:01:23,240 --> 00:01:29,810
So I repeat accuracy, which is precision of the foundational LM model.

15
00:01:30,380 --> 00:01:31,640
The cost.

16
00:01:32,860 --> 00:01:44,170
The latency, which is the speed and the privacy, meaning the ability to protect private data.

17
00:01:44,680 --> 00:01:45,340
Right.

18
00:01:45,640 --> 00:02:01,900
So what were the top a LM Foundation models used on a 2023 by artificial intelligence engineers.

19
00:02:01,900 --> 00:02:13,210
So by far, by far the top, uh, the most used uh foundation model was ChatGPT.

20
00:02:14,650 --> 00:02:22,510
By far it is a during the 2023 year it has been the most accurate.

21
00:02:24,010 --> 00:02:27,010
By far, and this has been the top criteria.

22
00:02:28,500 --> 00:02:30,780
Very far from ChatGPT.

23
00:02:31,290 --> 00:02:44,760
We had we had another LM models like Anthropic Cahir, palm, llama two Falcon, etc. the interesting

24
00:02:44,760 --> 00:02:56,940
thing is that these 2023 survey showed that 80% of artificial intelligence engineers are experimenting

25
00:02:56,940 --> 00:02:59,400
with more than one lm.

26
00:03:00,280 --> 00:03:10,480
So even when they finally decide to use ChatGPT in the process, they have check different, uh, LM

27
00:03:10,480 --> 00:03:11,380
models.

28
00:03:11,890 --> 00:03:25,630
This means that when we have a and the alternative LMS, uh, in the same basic, uh, levels of accuracy,

29
00:03:25,750 --> 00:03:33,880
uh, or and that, uh, ChatGPT is providing right now, probably we are going to have a much more alternatives

30
00:03:33,880 --> 00:03:41,920
in LM applications, but until now the top choice is being ChatGPT.

31
00:03:45,730 --> 00:03:54,970
In the next lesson, we are going to talk about the selection of the stack of tools when building an

32
00:03:54,970 --> 00:03:56,620
LM application.