WEBVTT

00:00.200 --> 00:01.720
Let me go and show you the link.

00:01.720 --> 00:02.080
Smith.

00:02.120 --> 00:02.640
Trace.

00:03.160 --> 00:05.640
So I first want to go and log in.

00:08.000 --> 00:10.800
Let me go to the react algorithm project.

00:12.760 --> 00:14.640
And we can see now the trace.

00:14.800 --> 00:18.680
So we have two LM calls and one tool execution.

00:18.840 --> 00:21.280
So let's start with the first LM call.

00:21.480 --> 00:25.240
And we can see now that this was the input.

00:25.240 --> 00:34.000
And the result was a tool call with the function of get text length with the text equals to dog as input.

00:34.400 --> 00:37.800
So we can see also the tool call ID here.

00:38.000 --> 00:44.280
Now one thing which I like about blacksmith, you can see now that the tools here was actually a call

00:44.320 --> 00:44.760
tool.

00:44.800 --> 00:51.240
So Langschmidt is able to do it because we propagate through the entire life cycle the tool call ID.

00:51.440 --> 00:53.960
So Linkchain knows also to track it with Dan Smith.

00:54.120 --> 01:00.080
So here we can see the tool execution of a the function of get text length.

01:00.320 --> 01:02.720
Now we can see the input.

01:02.720 --> 01:04.440
We can see the output.

01:04.480 --> 01:06.310
Let me go and show you.

01:06.470 --> 01:08.870
The last message to the LM.

01:09.030 --> 01:11.630
And notice the list of messages we sent.

01:11.670 --> 01:14.470
Here we send the input, which was the first message.

01:14.470 --> 01:16.430
The second message was the tool call.

01:16.710 --> 01:18.550
The third message was the tool result.

01:18.550 --> 01:20.230
And then we get the output.

01:22.150 --> 01:22.870
Amazing.

01:22.870 --> 01:26.070
Let me go back to the code and let me commit everything.

01:26.310 --> 01:29.270
So I opened Cloud code in the right here.

01:29.270 --> 01:34.430
And I'm actually going to ask Cloud Code to create a commit message for me.

01:34.670 --> 01:40.830
And I actually found cloud code commit messages very, very appealing because they're short, they're

01:40.830 --> 01:41.670
descriptive.

01:41.670 --> 01:43.630
And I really like the formatting here.

01:43.910 --> 01:46.030
So let me simply write the prompt.

01:46.030 --> 01:49.510
I'm going to give it a quick hint that this commit is about tool calling.

01:49.710 --> 01:52.950
And you can see it's going to run the changed files.

01:52.990 --> 01:54.430
Git status git diff.

01:54.470 --> 01:56.790
Going to see exactly what we changed.

01:56.790 --> 01:59.510
And it's going to draft now a commit message.

02:00.030 --> 02:03.670
So let me say that we want to track the main.py file.

02:03.950 --> 02:06.430
And boom we can see now the commit message.

02:06.470 --> 02:09.860
And I'm always amazed by how amazing AI eyes.

02:09.980 --> 02:11.060
And so check it out.

02:11.220 --> 02:16.220
So in the description it's called replace react algorithm with link chain tool calling approach.

02:16.580 --> 02:20.140
We remove the manual react prompt template in parsing logic.

02:20.380 --> 02:24.100
We implemented the binding tools to direct integration with LLM.

02:24.380 --> 02:25.500
And this is what amazes me.

02:25.540 --> 02:29.740
Replaces the agent scratchpad tracking with message based conversation flow.

02:29.900 --> 02:32.260
So this is this is simply amazing.

02:32.540 --> 02:37.020
And let me say that I want to proceed and I want to commit this message.

02:37.900 --> 02:40.180
So now it's going to run this command.

02:40.900 --> 02:46.700
And once it's finished now I can go and push everything to the remote repository.

02:49.100 --> 02:51.420
So let me go to the GitHub repo.

02:51.420 --> 02:56.540
And here you can see the exact diff and everything we just showed you in the video.

02:56.700 --> 03:01.540
And that will also be linking it in the courses and resources in the videos.

03:01.540 --> 03:02.220
Resources.

03:02.780 --> 03:05.060
So yeah that's pretty much about tool calling.

03:05.100 --> 03:06.620
Hope you enjoyed this video.

03:07.100 --> 03:08.660
Please let me know what you think.

03:08.820 --> 03:13.260
And this is the natural evolution of the react prompt.
