1
00:00:02,360 --> 00:00:02,880
Hi guys.

2
00:00:02,880 --> 00:00:03,800
Welcome to this session.

3
00:00:03,800 --> 00:00:08,280
So in this session we want to talk about ChatGPT tokens which you get to see over here.

4
00:00:08,560 --> 00:00:14,760
So tokens are you can consider like pieces of words which are being used and counted over here.

5
00:00:14,760 --> 00:00:20,800
So when you give a prompt specifically the tokens are generated and ChatGPT different versions have

6
00:00:20,800 --> 00:00:23,120
different limits token limits out there.

7
00:00:23,120 --> 00:00:33,600
For example, ChatGPT 3.5 had a token limit of 4096 tokens and ChatGPT four later on had 8000 tokens.

8
00:00:33,600 --> 00:00:38,360
And now that we have new versions of it, they are much higher number of tokens which we get over here.

9
00:00:38,560 --> 00:00:44,880
So how it works is whenever there is a prompt which you give to ChatGPT, the prompt will take some

10
00:00:44,880 --> 00:00:46,120
of the tokens from there.

11
00:00:46,120 --> 00:00:53,480
So let's say you give a really long prompt to ChatGPT 3.5, wherein it it uses up.

12
00:00:53,520 --> 00:00:57,480
Out of 4096 it uses up 4000 tokens.

13
00:00:57,480 --> 00:01:02,680
So now we are left with only 96 tokens for ChatGPT to respond back.

14
00:01:02,680 --> 00:01:10,080
So your input and ChatGPT output both are considered in the total token amount.

15
00:01:10,120 --> 00:01:11,800
The limit which we have got here.

16
00:01:11,960 --> 00:01:17,120
So that is why you might see certain times when you are having a long conversation with ChatGPT.

17
00:01:18,200 --> 00:01:23,920
At the later stage of it, the the responses might not be that accurate, might not be that sensible

18
00:01:24,200 --> 00:01:25,520
information which you are getting.

19
00:01:25,560 --> 00:01:31,320
So in such a scenario, the hack which you can think of is starting a new chat, or what you can do

20
00:01:31,320 --> 00:01:37,440
is you can go ahead and summarize the complete conversation, ask ChatGPT to summarize the whole conversation

21
00:01:37,440 --> 00:01:43,760
in a concise manner, and then copy that into a new chat and start from there again, so that you have

22
00:01:43,880 --> 00:01:46,920
the fresh number of tokens again generated for the new chat.

23
00:01:47,200 --> 00:01:54,600
So there are also different ways you can figure out how much tokens a particular prompt will will take

24
00:01:54,600 --> 00:01:55,040
up.

25
00:01:55,040 --> 00:01:59,800
So that also we have tools like a tokenizer which tool which you can use over here.

26
00:01:59,800 --> 00:02:06,360
So first let's look at how OpenAI defines tokens Uh, on their platform.

27
00:02:07,160 --> 00:02:11,240
So tokens can be thought of as pieces of words, which they have.

28
00:02:11,280 --> 00:02:16,560
So you can see one token is almost approximately equal to four characters in English.

29
00:02:16,920 --> 00:02:17,400
Okay.

30
00:02:17,440 --> 00:02:23,360
And 1 to 2 sentences becomes around approximately 30 tokens, one paragraph, approximately 100 tokens,

31
00:02:23,360 --> 00:02:24,520
and so on and so forth.

32
00:02:24,800 --> 00:02:29,840
So here you can read about the token limits, token pricing, even exploring tokens.

33
00:02:29,840 --> 00:02:34,360
So here you can see uh, every particular word gets a specific token.

34
00:02:34,360 --> 00:02:39,880
So for example my is 33666 color is 3124.

35
00:02:40,040 --> 00:02:42,280
Then red is 2266.

36
00:02:42,280 --> 00:02:45,080
Now if you look at period period is 13.

37
00:02:45,080 --> 00:02:48,080
So which is remains the same everywhere in the second one as well.

38
00:02:48,120 --> 00:02:49,720
Period is given as 13.

39
00:02:49,760 --> 00:02:58,000
However, if you look at red uh in lower cases is 2266, whereas red with uppercase is 2297.

40
00:02:58,360 --> 00:03:00,480
So like this it differs.

41
00:03:00,480 --> 00:03:05,120
And this is how uh you can see tokens are used up in our prompts.

42
00:03:05,160 --> 00:03:09,990
Now if you want to calculate, uh, a particular prompt will take up how much tokens.

43
00:03:09,990 --> 00:03:12,030
So you can use tokenizer over here.

44
00:03:22,190 --> 00:03:25,470
So you can see this particular sentence will take about seven tokens.

45
00:03:25,470 --> 00:03:26,790
Characters are 28.

46
00:03:28,310 --> 00:03:32,390
If we pick a bigger text, let's say we are picking up this big text.

47
00:03:39,790 --> 00:03:43,110
So in that case it's taking 81 tokens and characters are 371.

48
00:03:43,150 --> 00:03:44,950
Each of them have been color coded.

49
00:03:44,950 --> 00:03:46,710
Now in this manner you can understand.

50
00:03:47,230 --> 00:03:52,510
So this is the idea behind tokens which needs to be taken into consideration.

51
00:03:52,510 --> 00:03:58,350
So whenever you are using ChatGPT for different reasons, uh, keep this in mind at the back end of

52
00:03:58,350 --> 00:04:03,950
your mind, uh, so that you are aware about it and you can optimize accordingly so that you get better

53
00:04:03,950 --> 00:04:04,870
responses.

54
00:04:05,230 --> 00:04:09,790
Thank you so much, guys, for listening to this and I will see you in the next video.