WEBVTT

00:00.000 --> 00:01.980
-: Another thing you can do with large language models,

00:01.980 --> 00:05.430
is to specify a rough output length, and a desired format.

00:05.430 --> 00:07.807
In this example, we've got a prompt that says,

00:07.807 --> 00:09.630
"Generate two to three sentences

00:09.630 --> 00:12.360
about why learning prompt engineering is important."

00:12.360 --> 00:15.060
Notice how we've specified a desired length.

00:15.060 --> 00:16.710
Let's have a look at how this performs.

00:16.710 --> 00:19.110
And as we can see, we've got three sentences.

00:19.110 --> 00:20.760
Let's now test an alternative prompt.

00:20.760 --> 00:22.507
In this example we're gonna say,

00:22.507 --> 00:24.870
"Generate three large paragraphs

00:24.870 --> 00:27.630
about why learning prompt engineering is important."

00:27.630 --> 00:29.400
Okay, so this prompt is nearly finished up,

00:29.400 --> 00:30.780
but as you can see,

00:30.780 --> 00:33.840
through specifying a different desired length,

00:33.840 --> 00:36.780
I generate three large paragraphs versus generate

00:36.780 --> 00:38.250
two to three sentences,

00:38.250 --> 00:40.440
can have a massive impact,

00:40.440 --> 00:42.960
in terms of the response length you get back

00:42.960 --> 00:44.370
from a large language model.

00:44.370 --> 00:46.500
Consider this another one of the different types

00:46.500 --> 00:49.020
of parameters that you can put inside of a prompt,

00:49.020 --> 00:52.143
to choose the response length that suits your task.