WEBVTT 00:02.080 --> 00:02.920 Behind the scenes. 00:02.920 --> 00:08.000 Most tools like cursor use something which is called an agent. 00:09.400 --> 00:13.200 You can think of an agent like a smart helper that can take actions. 00:13.200 --> 00:19.320 For example, working with files, running commands in a terminal, writing code, fixing bugs, and 00:19.320 --> 00:21.720 even interact with external tools. 00:22.040 --> 00:25.520 But here's the catch the agent can do so much on its own. 00:25.680 --> 00:28.920 It needs to be connected to something more powerful. 00:29.480 --> 00:31.320 A large language model. 00:31.520 --> 00:33.880 And that is the brain behind the scenes. 00:33.880 --> 00:40.680 The LLM is what actually understands your words and generates responses, writes the code, which then 00:40.680 --> 00:43.240 the AI transforms into the project. 00:44.120 --> 00:49.880 In this lecture, I want to point out something very important about this large language models, and 00:49.880 --> 00:53.360 that is the fact that they are non-deterministic. 00:53.840 --> 00:55.280 So what do I mean by that? 00:55.320 --> 01:02.680 Well, it means that if you give the same input or prompt, you can still get a different answer each 01:02.680 --> 01:03.120 time. 01:03.560 --> 01:06.440 Throughout the course I'm going to provide you the prompts that I use. 01:06.440 --> 01:10.350 But you're going to see that what you're getting back is slightly different. 01:10.590 --> 01:11.910 But why does it happen? 01:12.270 --> 01:15.590 Imagine asking the same question to a human three times. 01:15.910 --> 01:20.030 You will not get the exact same words each time, right? 01:20.190 --> 01:22.390 And this is how llms work. 01:22.430 --> 01:26.670 Also, they are not robots that only say the same thing. 01:26.710 --> 01:30.990 They are a bit like people, a bit creative, a bit random. 01:31.550 --> 01:33.710 And this can happen for a few reasons. 01:33.990 --> 01:40.350 One of these reasons is something called temperature, which is a setting that controls how creative 01:40.390 --> 01:41.430 the model can be. 01:42.070 --> 01:44.590 A higher temperature makes it more random. 01:45.350 --> 01:50.590 A lower one makes it more focused and predictable, which is generally good for writing code. 01:51.150 --> 01:56.670 While this is a key concept in AI, most tools don't expose this as a setting, so don't go looking 01:56.670 --> 01:57.230 for it. 01:57.270 --> 01:59.350 But I think it's still important to know about it. 01:59.990 --> 02:05.830 Apart from the setting, each large language model has its own personality if you want to call it like 02:05.830 --> 02:09.590 that, because it has its own training, data and design. 02:09.790 --> 02:15.870 So even if we send the same prompt to two different models that are tuned, let's say, with the same 02:15.870 --> 02:16.310 temperature. 02:16.310 --> 02:21.590 For example, when answering that question, they will still give different responses and quite often 02:21.590 --> 02:23.230 they will be quite, quite different. 02:24.190 --> 02:29.150 But no, this small randomness that we're getting is still helpful. 02:29.150 --> 02:36.270 It can give us fresh ideas, alternative solutions, and can help us explore different ways of solving 02:36.310 --> 02:37.230 a problem. 02:37.230 --> 02:41.710 So this is something to keep in mind if you're not happy with the result that you're getting with the 02:41.710 --> 02:48.230 first try, you can just delete everything and start from scratch and get a different solution and see 02:48.230 --> 02:49.430 if you like that better. 02:49.710 --> 02:52.510 It's just rolling the dice once again. 02:53.590 --> 02:56.350 So this is just a heads up before we move on. 02:56.630 --> 03:02.350 You might try the same prompt I'm showing on the screen and get slightly different results or totally 03:02.350 --> 03:03.390 different results. 03:03.710 --> 03:05.750 And this is absolutely normal. 03:05.750 --> 03:09.470 It doesn't mean anything is broken, it is just how these tools work. 03:09.830 --> 03:13.230 You just need to continue the chat based on the response you get. 03:13.670 --> 03:16.670 Each conversation is slightly different.