1
00:00:03,380 --> 00:00:03,740
Okay.

2
00:00:03,740 --> 00:00:13,850
So as we were saying in the article, Long Chain has described the full professional LM application

3
00:00:13,850 --> 00:00:21,110
development cycle and they have identified three main main phases.

4
00:00:21,110 --> 00:00:22,460
Prototyping.

5
00:00:23,140 --> 00:00:26,980
When we develop the first prototype of our application.

6
00:00:27,340 --> 00:00:28,990
Beta testing.

7
00:00:28,990 --> 00:00:36,970
When we test our prototype with real users and get their feedback to make improvements and production.

8
00:00:36,970 --> 00:00:42,190
When we monitor the performance of our application and keep improving it.

9
00:00:43,420 --> 00:00:50,770
Let's see the details that Land Chain shared about each of these, uh, phases.

10
00:00:51,560 --> 00:00:52,730
About prototyping.

11
00:00:52,730 --> 00:01:00,650
They say it involves quick experimentation between prompts, model types, retrieval strategy, and

12
00:01:00,650 --> 00:01:01,970
other parameters.

13
00:01:02,630 --> 00:01:11,600
The ability to rapidly understand how the model is performing and debug where it is failing is incredibly

14
00:01:11,600 --> 00:01:13,940
important for this phase.

15
00:01:15,440 --> 00:01:18,680
What do they say about the beta testing?

16
00:01:19,340 --> 00:01:20,180
Face.

17
00:01:20,810 --> 00:01:29,120
Beta testing allows developers to collect more data on how their LM applications are performing in real

18
00:01:29,120 --> 00:01:30,620
world scenarios.

19
00:01:31,130 --> 00:01:38,540
In this phase, it is important to develop an understanding for the types of input the app is performing

20
00:01:38,540 --> 00:01:44,570
well or poorly on, and how exactly it is breaking down in those cases.

21
00:01:45,320 --> 00:01:51,980
Both feedback collection and run annotation are critical for this workflow.

22
00:01:52,280 --> 00:02:00,410
This will help in creation of test cases that can help track regressions or improvement and development

23
00:02:00,410 --> 00:02:03,350
of automatic evaluations.

24
00:02:04,100 --> 00:02:07,340
What do they say about the production phase?

25
00:02:08,759 --> 00:02:18,300
Closely inspecting key data points, growing benchmarking data sets, annotating traces, and drilling

26
00:02:18,300 --> 00:02:20,400
down into important data.

27
00:02:20,400 --> 00:02:28,230
In Trace View are workflows you will also want to do once your application hits production.

28
00:02:29,100 --> 00:02:36,960
However, especially at the production stage, it is crucial to get a high level overview of application

29
00:02:36,960 --> 00:02:42,990
performance with respect to latency, cost and feedback scores.

30
00:02:43,020 --> 00:02:49,560
This ensures that it is delivering the desirable results at the scale.

31
00:02:50,590 --> 00:03:01,570
This is how the launching team describes a the full professional LM application development cycle on

32
00:03:01,570 --> 00:03:02,710
a high level.

33
00:03:02,890 --> 00:03:09,670
In the next lessons, we are going to see the details of these phases and especially the main challenges

34
00:03:09,670 --> 00:03:18,670
that developer teams are finding and how long Smith is helping us to solve these problems.