WEBVTT

00:00.180 --> 00:03.300
-: Certain tasks are best described in a step-by-step manner.

00:03.300 --> 00:05.940
It's possible for us to tell a language model,

00:05.940 --> 00:08.820
such as GPT-4 or ChatGPT,

00:08.820 --> 00:10.680
a step-by-step approach,

00:10.680 --> 00:13.620
which we have predetermined or predefined.

00:13.620 --> 00:16.440
Let's have a look at how this works inside of an example.

00:16.440 --> 00:18.220
In this prompt, we've specified two steps

00:18.220 --> 00:20.520
that we'd like ChatGPT to follow.

00:20.520 --> 00:22.680
The first step, Step 1, is:

00:22.680 --> 00:24.540
The user will hand over a chunk of text

00:24.540 --> 00:26.340
enclosed in triple quotes.

00:26.340 --> 00:28.770
So we're gonna enclose our text in triple quotes.

00:28.770 --> 00:31.680
Your task is to condense this text into a single sentence

00:31.680 --> 00:34.260
starting with the phrase summary.

00:34.260 --> 00:37.830
Step two, translate the summarization you made in step one

00:37.830 --> 00:40.170
into Spanish, leading the translation

00:40.170 --> 00:42.480
with the phrase, translation.

00:42.480 --> 00:43.560
And then we've got this saying,

00:43.560 --> 00:45.180
I love learning about prompt engineering.

00:45.180 --> 00:47.130
It's pretty interesting, let's keep going.

00:47.130 --> 00:51.300
Chat GPT has effectively done both steps sequentially.

00:51.300 --> 00:54.270
Firstly, we've got one sentence we've summary.

00:54.270 --> 00:56.087
The user expresses interest in enthusiasm in learning

00:56.087 --> 01:00.240
about prompt engineering, and we've also got a translation.

01:00.240 --> 01:04.860
Also, notice that it utilized our prefixes, summary,

01:04.860 --> 01:05.940
and translation.

01:05.940 --> 01:07.980
So if you have a task that you think needs

01:07.980 --> 01:11.662
to be done in a series of steps, you could specify

01:11.662 --> 01:13.620
the steps beforehand,

01:13.620 --> 01:15.330
insert these into a prompt,

01:15.330 --> 01:17.130
and that would encourage the language model

01:17.130 --> 01:18.963
to follow that chain of thought.