WEBVTT

00:01.760 --> 00:02.480
Friends.

00:02.510 --> 00:05.390
I am absolutely exhausted.

00:05.390 --> 00:09.620
I am exhausted and a little tiny bit traumatized.

00:09.950 --> 00:13.520
And you are somewhat to blame for this, as you will discover.

00:13.550 --> 00:16.760
You will discover in a moment you will find out why.

00:16.790 --> 00:21.050
But first, let's talk about what we have in store for today.

00:21.500 --> 00:25.670
So we are going to be raising the bar on our baselines.

00:25.670 --> 00:29.360
I'm just going to take a moment to remind you what you can already do.

00:29.390 --> 00:36.230
Of course, generating text and generating code from combinations of frontier models with AI assistants,

00:36.230 --> 00:44.240
with tools, and also open source systems using hugging face Transformers library, you can use Lang

00:44.240 --> 00:51.110
chain to build a rag pipeline, and now you can also curate data and you can curate it.

00:51.110 --> 00:52.700
Finally, if I may say so.

00:52.700 --> 00:57.740
And you can also make a baseline model using some rather foolish techniques.

00:57.740 --> 01:04.040
But then using traditional machine learning linear regression, including both feature engineering and

01:04.040 --> 01:11.030
bag of words, and then onto more sophisticated techniques using word two vec and then adding in support

01:11.030 --> 01:14.270
vector machines and then random forests.

01:14.330 --> 01:16.250
Quite a trek it's been.

01:16.250 --> 01:17.750
So today.

01:17.780 --> 01:23.510
Today we are going to now take the framework we put together and put it against frontier models.

01:23.510 --> 01:30.080
And this will sort of really capture all of the steps it takes to take a proper business problem, um,

01:30.140 --> 01:35.750
understand the data and then present it to frontier models and compare their performance.

01:35.750 --> 01:41.870
So it's an exciting moment for us, and we will get right to it.

01:41.930 --> 01:46.130
And I will see you over at JupyterLab in a moment.