WEBVTT 00:00.280 --> 00:06.080 And they then some time ago released a new feature, which is absolutely outstanding because it makes 00:06.080 --> 00:08.880 our agents 3 to 5 times better. 00:08.880 --> 00:15.280 They are more productive, accurate, and even we can save a lot of time by applying this simple feature. 00:15.480 --> 00:17.760 And I'm talking about the think node. 00:18.200 --> 00:21.440 So in general in this video will cover everything about it. 00:21.440 --> 00:23.800 And also two very important use cases. 00:23.960 --> 00:30.400 So the first one will be about increasing the chances of your agent does exactly what you want. 00:30.440 --> 00:35.800 Because we know that for an example, when we run some prompt, sometimes it produces the response which 00:35.800 --> 00:36.640 is just bad. 00:36.680 --> 00:38.080 It misunderstand that. 00:38.080 --> 00:41.080 And a lot of like more issues we have here. 00:41.120 --> 00:46.800 Nevertheless, with this feature we can increase our chances that all right we get what we want. 00:46.840 --> 00:54.160 And the second use case is to turn our agent to speak more like a human, not actually as an AI. 00:54.840 --> 00:57.280 Um, so with all that being said, let's get started. 00:57.320 --> 01:00.080 And firstly, I can show you how it works. 01:00.080 --> 01:02.120 So how think note works. 01:02.600 --> 01:10.080 And we've got this simple, um, a agent that is connected to OpenAI and the think node, and we trigger 01:10.120 --> 01:13.960 this, um, agent with the message so I can open the chat. 01:14.480 --> 01:17.840 And here I prepared some message I can paste even. 01:17.840 --> 01:20.080 Let me show you this prompt here. 01:20.080 --> 01:21.760 So it's nicely visible. 01:21.960 --> 01:26.320 So a government wants to reduce car traffic in major cities by introducing a new law. 01:26.360 --> 01:32.680 Every citizen must choose between owning a private car or using public transportation, but not both. 01:32.720 --> 01:39.480 Ride sharing apps and taxis will be still allowed, but owning a car disqualifies you from using subsidized 01:39.480 --> 01:41.720 public transport, and vice versa. 01:41.760 --> 01:46.760 Your task is to evaluate the policy as a smart, long term solution. 01:46.760 --> 01:51.600 So step by step, identify key assumptions behind the policy, predict likely outcomes and different 01:51.600 --> 01:58.600 types of cities dense, sub, urban or rural, and list at least three unintended intended consequences 01:58.600 --> 02:03.440 this policy might cause suggest refinements or adjustments to the policy to make it more realistic and 02:03.440 --> 02:03.840 fair. 02:03.880 --> 02:07.080 Conclude whether the revised policy is better than doing nothing. 02:07.600 --> 02:09.080 So now I can copy that. 02:09.120 --> 02:14.240 Go over to my to my na den and here just paste it inside chat. 02:14.480 --> 02:17.360 And now we can see how it works. 02:17.520 --> 02:21.600 So we've got our new message agent and open AI. 02:22.440 --> 02:26.680 And actually it will use the think note to think about it. 02:26.680 --> 02:31.920 So in the common situation we've got our agent that uses only open AI. 02:31.960 --> 02:32.320 All right. 02:32.360 --> 02:35.080 So we've got you know like the input and output. 02:35.280 --> 02:39.560 So this is the entire process that goes with the LMS actually. 02:39.560 --> 02:40.680 And other videos. 02:40.760 --> 02:43.000 Um we've got explained how it works. 02:43.040 --> 02:49.120 However simply it takes the input, it predicts actually the next words and it creates the response. 02:49.120 --> 02:49.560 All right. 02:49.600 --> 02:51.520 Based on some specific database. 02:51.640 --> 02:56.800 Nevertheless, with the think note, it actually thinks like a real human. 02:56.840 --> 02:58.160 So let me explain that. 02:58.200 --> 03:00.240 Firstly we've got the input, as you can see here. 03:00.280 --> 03:00.840 All right. 03:00.840 --> 03:01.760 We've got the input. 03:01.760 --> 03:02.440 So human. 03:02.480 --> 03:04.520 A government wants to reduce car traffic. 03:04.520 --> 03:05.680 So this is our prompt. 03:05.880 --> 03:12.800 Then what it did instead of just going like straight forward to creating for us the response it used 03:12.800 --> 03:13.600 the think node. 03:13.600 --> 03:17.200 And now you can see it identifying key assumptions behind the policy. 03:17.320 --> 03:23.000 And one limiting the ownership of private cars will significantly reduce traffic congestion in urban 03:23.000 --> 03:23.360 areas. 03:23.360 --> 03:28.440 Citizens will easily adapt to using public transportation if they don't have a private car. 03:28.680 --> 03:33.080 So it just took the prompt and it thought for some time. 03:33.400 --> 03:36.480 Again, we've got another thing actually. 03:36.480 --> 03:37.000 Note. 03:37.000 --> 03:40.320 So predicting likely outcomes and different types of cities. 03:40.320 --> 03:47.400 So here we had the policy there the cities there we have listing unintended consequences. 03:48.200 --> 03:54.280 So here as you can see like the thinking process is as we would have done it as humans. 03:54.480 --> 03:56.870 So the think note is so so nice. 03:57.110 --> 03:59.230 And here is actually the comparison. 03:59.390 --> 04:04.550 Think note inside Nadhan is like the reasoning model right. 04:04.590 --> 04:08.150 So you've got reasoning models like Deep seek and R1. 04:08.390 --> 04:11.790 And here for an example OpenAI or one. 04:11.790 --> 04:15.430 So these are reasoning models and it works like pretty the same. 04:15.470 --> 04:15.790 All right. 04:15.790 --> 04:18.470 Because we've got yeah this think node. 04:18.470 --> 04:27.390 So we just turn our typical LLM which is this one OpenAI to just the reasoning model kind of and there. 04:27.430 --> 04:29.590 All right let's let's preview what we have else. 04:29.630 --> 04:31.350 Firstly we have the input three. 04:31.710 --> 04:34.150 Um no three actually steps for thinking. 04:34.390 --> 04:39.030 The fourth step is actually creating, you know, creating the output. 04:39.430 --> 04:44.870 The fourth step we've got our input and that is refined for our OpenAI model. 04:44.990 --> 04:46.070 So here what I did. 04:46.110 --> 04:49.150 It took our, you know, data from the prompt we provided. 04:49.150 --> 04:54.830 And here also it added there um the outputs from thinking notes. 04:54.830 --> 05:00.750 So here we've got three actually, you know, we've got key assumptions behind the policy, um, outcomes 05:00.750 --> 05:02.270 and different types of cities. 05:02.270 --> 05:04.510 And here, um, consequences. 05:04.750 --> 05:11.470 So what I did again actually took our input, thought, um, about some circumstances, updated the 05:11.470 --> 05:12.990 prompt here. 05:12.990 --> 05:15.150 Again, it thought for the next time. 05:15.390 --> 05:18.390 Then again, it updated the prompt with this thought. 05:18.510 --> 05:20.030 Again it thought, all right. 05:20.030 --> 05:24.030 And at the end we've got the output after all of this data. 05:24.030 --> 05:30.670 So it collected the data here and at the end like used it um in our output. 05:30.710 --> 05:37.990 And it's absolutely outstanding because it thinks it allows actually think note allows our agent to 05:38.030 --> 05:38.470 think. 05:38.510 --> 05:44.790 And by this way we can avoid a lot of different mistakes in the future because it refines it, recalls 05:44.790 --> 05:46.270 some data, it checks. 05:46.310 --> 05:46.750 All right. 05:46.790 --> 05:50.790 Um, if this data was correct, if I was correct or not. 05:51.230 --> 05:51.590 Yeah. 05:51.590 --> 05:53.270 So it works by this way. 05:53.550 --> 05:57.350 And by the way, the think node was inspired by anthropic article. 05:57.350 --> 06:03.190 So here they introduced the think tool enabling cloud to stop and think and complex tool use situations. 06:03.590 --> 06:07.390 So here um, actually you've got we won't read this entire article. 06:07.430 --> 06:10.590 However we'll yeah like take the most important parts. 06:10.630 --> 06:16.590 So here, as we continue to enhance clouds complex problem solving abilities, we've discovered a particularly 06:16.590 --> 06:22.470 effective approach, a think tool that creates dedicated space for structures thinking during complex 06:22.470 --> 06:23.070 tasks. 06:23.230 --> 06:29.550 So imagine, um, actually you are solving some complex math problem and you've got the paper and you 06:29.550 --> 06:32.790 can like, you know, you can like type your thoughts on that. 06:32.790 --> 06:39.830 So as you saw in this entire example what it did, again, it used the thinking process to just create 06:39.830 --> 06:40.910 for us some data. 06:40.910 --> 06:44.070 And afterwards it used that and our final response. 06:44.070 --> 06:46.710 So it makes a lot of sense here. 06:46.750 --> 06:52.030 What we can read this simple yet powerful technique, which as we'll explain below, is different from 06:52.030 --> 06:52.390 clouds. 06:52.390 --> 06:52.910 All right. 06:52.910 --> 06:54.750 So we are not interested about it. 06:54.790 --> 06:57.630 Um, here what is the thing tool? 06:57.630 --> 07:01.670 With the thing tool, we are giving cloud ability to include an additional thinking step. 07:01.670 --> 07:05.470 And again this is analogy for the think tool inside an addon. 07:05.670 --> 07:07.670 So um yeah it's about it. 07:07.950 --> 07:13.790 Um, additional thinking step complete with its own design space as part of getting to its final answer. 07:13.990 --> 07:15.750 The thing tool is for cloud. 07:15.750 --> 07:21.350 Once it starts generating a response, to add a step to stop and think about whether it has all of the 07:21.350 --> 07:23.230 information it needs to move forward. 07:23.230 --> 07:29.350 This is particularly helpful when performing long chains of tool calls or in long, multi-step conversations 07:29.350 --> 07:30.350 with the user. 07:30.390 --> 07:32.430 That's why here we had the case. 07:32.430 --> 07:38.270 So, you know, we had um, yeah, very complex problem that we would need to solve. 07:38.470 --> 07:45.110 And in these circumstances this thing tool and think node is just perfect because it breaks the problem 07:45.110 --> 07:46.150 into smaller pieces. 07:46.150 --> 07:53.430 It thinks it just, you know, as I said, the analogy to the paper and solving, you know, solving 07:53.430 --> 07:54.630 very complex problem. 07:54.630 --> 07:59.510 You just type your thoughts on paper and then you just refine your answer and do other stuff. 08:00.830 --> 08:05.270 This makes the thing tool more suitable for cases where cloud does not have all of the information needed 08:05.270 --> 08:11.710 to formulate its response from the user query alone, and where it needs to process external information. 08:12.510 --> 08:17.950 Also, in this article, you've got the recommended uses for um think um tool. 08:18.270 --> 08:19.630 It's about the cloud. 08:19.670 --> 08:23.670 However also it's yeah you know performs when it comes to denim. 08:23.710 --> 08:29.030 So here when to use the think tool tool output analysis when clouds needs to carefully process the output 08:29.030 --> 08:33.830 of previous tool calls before acting and might need to backtrack in its approach. 08:33.910 --> 08:41.070 So also if you have actually, let's say a voice agent and let's say you are, um, making appointments, 08:41.070 --> 08:43.310 you can use this tool to actually see. 08:43.350 --> 08:43.710 All right. 08:43.750 --> 08:49.630 So actually think if for an example that we've got the data which is available or something else. 08:49.630 --> 08:54.430 When you create some complex text, like a lot of different use cases, we won't cover that in this 08:54.430 --> 08:54.990 video. 08:55.030 --> 09:01.070 However, um, you know, after you know how to use it, you can implement that in many, many cases. 09:01.230 --> 09:06.310 So here, policy heavy environments when cloud needs to follow detailed guidance and verify compliance. 09:06.590 --> 09:08.030 Here decision making. 09:08.150 --> 09:14.670 So also decision making is a great use case because it it needs to like think a lot break the problem 09:14.670 --> 09:15.870 into smaller pieces. 09:15.870 --> 09:19.990 So when each action builds on previous ones and mistakes are costly. 09:20.270 --> 09:28.150 And that's why I said at the beginning that we, um, we created like this, um, sample is just actually 09:28.190 --> 09:34.510 it allows our agent to be more accurate, to be more precise, and to make just less mistakes. 09:34.510 --> 09:41.830 Because if it can think all right, if it thinks it's just better, it's simpler, it's it's better, 09:41.830 --> 09:43.110 it's more accurate. 09:43.510 --> 09:46.510 So it's very important implementation best practices. 09:46.510 --> 09:49.230 So here you've got strategic prompting here. 09:49.270 --> 09:51.550 Um, complex guidance and system prompts. 09:52.190 --> 09:55.710 And they're even for what you don't need to use think tool. 09:55.750 --> 10:00.590 So for sure if you have like simple cases right, you need to create some piece of the text that doesn't 10:00.590 --> 10:06.150 require like thinking like, you know, like something very sophisticated. 10:06.190 --> 10:07.190 Don't use this tool. 10:07.230 --> 10:13.550 Nevertheless, in most cases and in most automations, because you've got, um, for sure, you've got, 10:13.710 --> 10:17.910 um, these automations that are very advanced, you would like to use think tool. 10:18.150 --> 10:22.150 It makes your agents just better, more effective, like on the steroids. 10:23.310 --> 10:26.950 Now, let me show you how we can add the think tool and to your agent. 10:26.950 --> 10:34.350 And it's simple as I'm clicking on this plus um, under the tool and here just selecting the think so. 10:34.750 --> 10:36.550 So just search for the think tool. 10:36.710 --> 10:40.270 And here you've got it actually it has the prompt. 10:40.270 --> 10:45.110 So we can actually um adjust the behavior of the thinking note. 10:45.150 --> 10:47.230 However I will show you that in a while. 10:48.430 --> 10:53.950 And now it will be nicely visible on this example, because I will show you how to create an agent that 10:53.950 --> 10:59.750 speaks more naturally, like a human, not a robot, which is nice for a lot of circumstances. 10:59.750 --> 11:05.350 For example, if you create a piece of text, article, blog post, or even social media content. 11:05.350 --> 11:11.830 All right, so here inside the AI agent instructions, inside the prompt I've got here, I'm actually 11:11.830 --> 11:13.310 let me show you overview. 11:13.350 --> 11:14.790 You're a helpful assistant. 11:14.790 --> 11:20.830 And the instructions when you create any piece of written content you use a think node to make it human 11:20.830 --> 11:21.190 like. 11:21.230 --> 11:23.350 And then I provided the variable. 11:24.390 --> 11:26.790 So they're also inside the think node. 11:26.790 --> 11:30.790 As I said we can change the behavior of this node. 11:30.790 --> 11:34.190 So there I've cut this very long um actually prompt. 11:34.550 --> 11:39.270 And then you can actually copy from our file that will be in the resources of this material. 11:39.270 --> 11:42.750 So here you've got think node instructions and we've got the overview. 11:42.870 --> 11:48.500 So this tool should reflect on a message from a user and respond to it in a more natural, clear and 11:48.500 --> 11:49.260 human way. 11:49.300 --> 11:53.220 It should feel like a smart assistant taking a moment to think before saying something. 11:53.260 --> 11:57.460 And here we've got the personality and tone, human like language examples. 11:57.700 --> 11:58.060 All right. 11:58.100 --> 12:01.060 So what to use and suggest that vocabulary. 12:01.300 --> 12:02.060 And so there. 12:02.100 --> 12:08.460 For an example instead of slightly a bit and seems fix and words for example to avoid. 12:08.460 --> 12:09.020 All right. 12:09.060 --> 12:11.420 And now you will see that in action. 12:11.580 --> 12:16.780 So I've got here let's open the chat and let's see what we have. 12:16.940 --> 12:18.900 I've got some article prepared. 12:19.300 --> 12:20.860 Um so let me copy that. 12:20.860 --> 12:22.060 Let's paste it here. 12:22.420 --> 12:24.900 And in general also let me paste it here. 12:24.900 --> 12:27.820 So I will show you what we have in the current digital landscape. 12:27.860 --> 12:33.860 It is of utmost importance of enterprises to commence the implementation of artificial intelligence 12:33.860 --> 12:38.980 methodologies to facilitate improve, improved operation throughput. 12:39.020 --> 12:42.900 Organizations that endeavor to utilize cutting edge systems. 12:43.060 --> 12:46.740 So you can see this piece of the content is very rough. 12:46.780 --> 12:50.420 I can say to read or even it's very robotic. 12:50.660 --> 12:55.700 However, with, um, you know, with this agent, we've got the output. 12:55.940 --> 12:59.180 So here we can go right there and click on locks. 12:59.740 --> 13:00.460 Um, actually. 13:00.500 --> 13:00.900 All right. 13:00.900 --> 13:01.620 You see that? 13:02.420 --> 13:03.900 So here we've got the instructions. 13:03.900 --> 13:09.460 Think so in the current it took actually some pieces of the content and it refined that. 13:09.460 --> 13:10.220 All right. 13:10.500 --> 13:12.180 So let's see. 13:12.220 --> 13:12.620 Yeah. 13:12.820 --> 13:15.060 And here we've got our output. 13:15.140 --> 13:20.900 In today's digital world it's really important for businesses to start using AI to boost how they operate. 13:20.900 --> 13:26.700 Companies that jump on the new tech are often the ones that see big improvements in how well they do. 13:26.900 --> 13:29.220 So it's much more human like. 13:29.500 --> 13:33.220 It's just 100% better because the market is changing. 13:33.220 --> 13:38.780 Companies need to take a good look at their old systems to see if moving to an AI setup makes sense. 13:38.860 --> 13:40.620 So I wouldn't even. 13:40.780 --> 13:46.420 I'm distinguished Distinguish like this text, which is written by eye from a human. 13:46.460 --> 13:47.060 All right. 13:47.260 --> 13:48.500 And that wrote it. 13:48.500 --> 13:49.780 So it's simple as that. 13:49.780 --> 13:54.260 Actually you can use also this think note to make the text human like. 13:54.700 --> 14:00.020 And I think it's useful for especially for written content when you want to get, you know, like the 14:00.020 --> 14:07.060 best piece of text without any, you know, and the weird elements and even, um, all of us know like 14:07.100 --> 14:09.740 AI for now, llms are not that perfect. 14:09.740 --> 14:17.860 Sometimes it produces for us bad results, especially if we are using that a lot and we can occur the 14:17.860 --> 14:21.060 situations where we just get a bad response. 14:21.060 --> 14:26.980 I can say one per five responses and it's very frustrating and by this way you can avoid it. 14:26.980 --> 14:28.700 So you don't need to refine something. 14:28.700 --> 14:34.660 You can just more trust your AI agent because as I said at the beginning, it's more accurate, just 14:34.660 --> 14:35.180 better. 14:35.180 --> 14:38.540 And now even you can type, you know, like create better content. 14:38.940 --> 14:40.980 Um, with the thinking note.