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.
 
 

148 lines
4.0 KiB

WEBVTT
00:01.010 --> 00:01.580
Well.
00:01.580 --> 00:02.330
Hello there.
00:02.360 --> 00:04.550
Look, I know what you're thinking.
00:04.580 --> 00:07.070
You're thinking I peaked too early.
00:07.100 --> 00:09.950
Last week was an amazing climax.
00:09.950 --> 00:13.430
We built a model that massively outperformed the frontier.
00:13.460 --> 00:15.290
It can't get better than that.
00:15.320 --> 00:17.000
Well, I've got news for you.
00:17.090 --> 00:18.860
It's going to get better than that.
00:18.890 --> 00:22.670
We are actually going to have the best week yet.
00:22.670 --> 00:24.680
And it's just getting started.
00:24.860 --> 00:29.960
So this week, this week it's about reaching the pinnacle.
00:29.960 --> 00:33.770
It's about becoming a master of LM engineering.
00:33.770 --> 00:34.970
And we're going to do it.
00:35.000 --> 00:36.680
We're going to have a blast.
00:36.680 --> 00:38.510
It's going to be epic.
00:38.870 --> 00:44.030
Uh, we are going to build a multi-agent framework.
00:44.030 --> 00:48.290
We're going to build a solution that is going to be really, really awesome.
00:48.290 --> 00:54.740
And yeah, it's going to do something profound and it's going to do it using everything that we've been
00:54.740 --> 00:56.840
building for the last eight weeks.
00:57.140 --> 01:02.750
Uh, for today, though, we're going to be using modal, the serverless platform for AI, so that we
01:02.750 --> 01:09.390
can deploy the same model that we built last week, our proprietary Three specialized alum, we're going
01:09.390 --> 01:18.120
to be able to deploy it behind a serverless API on the cloud, and we're going to build the first agent
01:18.150 --> 01:18.930
of many.
01:18.960 --> 01:24.390
The first of the seven agents that are going to be part of the solution that we are building this week
01:24.510 --> 01:27.150
to solve a juicy business problem.
01:27.150 --> 01:35.550
So without further ado, let's get into week eight, the final week of mastering LLM engineering.
01:35.790 --> 01:40.110
And of course, it wouldn't be the start of a week without this picture.
01:40.110 --> 01:44.640
For the last time, let's look at the journey that you've been on.
01:44.670 --> 01:54.210
You started or seven weeks ago, uh, you first uh, we we spent some time looking at frontier models,
01:54.330 --> 02:00.150
uh, looking at asking questions, asking how many times that A appeared in this sentence, if you can
02:00.150 --> 02:03.540
remember that, uh, in week two, we built UIs.
02:03.540 --> 02:05.850
We experienced gradio for the first time.
02:05.850 --> 02:08.520
We worked with agent ization in a light way.
02:08.520 --> 02:10.080
It's going to get a lot heavier.
02:10.200 --> 02:16.340
Uh, and we played around with Multi-modality in week three, we went open source with hugging face
02:16.340 --> 02:19.760
with tokenization, with with models.
02:20.150 --> 02:26.930
In week four, we built that amazing code generation tool that rewrote Python a, C plus, plus.
02:26.930 --> 02:32.000
And in doing so, we explored how you select the right LLM to solve a business problem.
02:32.060 --> 02:37.760
In week five, we tackled Rag using Lang Chain and built a question answering project.
02:37.790 --> 02:39.890
Maybe you built something bigger as well.
02:39.920 --> 02:41.540
In week six.
02:41.540 --> 02:48.380
Week six was when we started on training and we fine tuned a frontier model and week seven.
02:48.410 --> 02:55.460
Of course, week seven we've just come from was when we beat that frontier model using Cultura fine
02:55.460 --> 03:00.410
tuning to make our own Verticalized specialized LLM.
03:00.890 --> 03:03.650
And I loved it and I hope you did too.
03:03.680 --> 03:10.430
And so this week, week eight, the culmination of everything that's come before is when we have a finale.
03:10.460 --> 03:18.860
We must LLM engineering and we're going to do it by focusing on Agentic AI and building an agent platform.
03:19.520 --> 03:21.350
Let's talk more about that next time.