You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 

118 lines
3.6 KiB

WEBVTT
00:00.110 --> 00:06.500
So I realized that day one of week one has been a pretty long day, and I assure you that the other,
00:06.500 --> 00:09.500
generally speaking, the other days won't be as long.
00:09.560 --> 00:14.930
We had some foundational work to do to get the environments up and running, and hopefully that you're
00:14.930 --> 00:19.730
happy that we've got there and you're feeling satisfied that we ran our first big project.
00:19.730 --> 00:27.200
As a quick recap of what we got done at the very beginning, seems like an age ago we used to run LMS
00:27.200 --> 00:33.500
open source LMS locally on your box, running them to generate content.
00:33.770 --> 00:41.540
Then we set up the environment, and then we used open AI in the cloud to make a call to frontier models
00:41.540 --> 00:42.560
to GPT four.
00:42.800 --> 00:49.670
Mini was the model we used to generate text there, and obviously we're using here a closed source model
00:49.670 --> 00:53.300
that is maybe 1000 or 10,000 times larger.
00:53.540 --> 00:59.420
We pay a small price for that in the form of a fraction of a cent, but we do have to to pay to use
00:59.420 --> 00:59.750
that.
00:59.750 --> 01:06.500
But what we get back is much richer in quality than using a small local one.
01:06.890 --> 01:12.590
Um, we learned how to distinguish between a system prompt and a user prompt just at a high level.
01:12.590 --> 01:15.710
We'll do a lot more on that, of course, in the coming days.
01:15.770 --> 01:21.860
Uh, system prompt setting the tone, the context of the conversation, the user prompt, which is the
01:21.860 --> 01:23.240
conversation itself.
01:23.270 --> 01:24.730
We used it for the opener.
01:24.760 --> 01:28.030
Later, we'll be using it for many rounds of conversation.
01:28.030 --> 01:35.320
And then most importantly, we applied this to the field of summarization and a critical use case that
01:35.320 --> 01:39.550
comes up so many times it's applicable to many different problems.
01:39.550 --> 01:44.260
It's something that I hope you'll find ways to use this in your day job, in what you do already.
01:44.260 --> 01:49.090
And if not, then certainly you should be able to find personal projects that you could come up with
01:49.090 --> 01:50.440
where you could apply this.
01:50.440 --> 01:53.110
And I'm really excited to see what people come up with.
01:53.530 --> 01:55.480
So that's what we got done.
01:55.480 --> 02:02.950
And would you believe we are already 2.5% through the course on the way to being an LLM engineering
02:03.190 --> 02:07.000
expert, so it's already progress has been made.
02:07.030 --> 02:12.460
Tomorrow we're going to talk about what really is that journey like what are the steps.
02:12.460 --> 02:16.600
So you have a clear sense of what's what's to be done, set you up for success.
02:16.600 --> 02:18.910
And then we'll do some, some, some content.
02:18.910 --> 02:22.870
We'll talk about what are the leading frontier models and the different ways to use them.
02:22.870 --> 02:28.900
And we'll also do some quick lab work, something I promised you that people who would prefer not to
02:28.930 --> 02:31.630
fork out dollars to OpenAI.
02:31.660 --> 02:37.360
I'm going to show you how we could use Olama as an alternative with the same code that we just wrote,
02:37.360 --> 02:43.870
calling Olama running locally instead of calling out to the frontier model on the cloud.
02:43.870 --> 02:46.030
So we'll do that tomorrow too.
02:46.060 --> 02:47.500
Very much looking forward to it.
02:47.500 --> 02:48.670
And I will see you then.