1 00:00:00,000 --> 00:00:01,260 -: Without further ado, 2 00:00:01,260 --> 00:00:04,380 let's hop into this prompt engineering section. 3 00:00:04,380 --> 00:00:05,550 So what I want you to do first 4 00:00:05,550 --> 00:00:08,310 is download this Prompt Engineering Guide 5 00:00:08,310 --> 00:00:11,520 that Benza and I put together to give you all the resources 6 00:00:11,520 --> 00:00:13,440 that you're gonna need for this section. 7 00:00:13,440 --> 00:00:16,620 So you can download it now in the Resources tab. 8 00:00:16,620 --> 00:00:19,560 Now, this guide is an overview of prompt engineering. 9 00:00:19,560 --> 00:00:22,380 What I wanna do in these lectures is just read through this 10 00:00:22,380 --> 00:00:23,790 so that you can follow along 11 00:00:23,790 --> 00:00:26,790 and you can use this as a complete GPS 12 00:00:26,790 --> 00:00:29,190 and cheat sheet for all your future prompts. 13 00:00:29,190 --> 00:00:30,630 So first, the overview. 14 00:00:30,630 --> 00:00:33,420 Like I mentioned, this prompt engineering doc is designed 15 00:00:33,420 --> 00:00:35,970 to optimize your ideation process 16 00:00:35,970 --> 00:00:38,580 and enhance the quality of your output. 17 00:00:38,580 --> 00:00:41,910 So this document is comprised of two main sections. 18 00:00:41,910 --> 00:00:43,470 There's a simplified section, 19 00:00:43,470 --> 00:00:45,810 and then there's an advanced prompting section. 20 00:00:45,810 --> 00:00:49,140 The simple section is ideal for creating quick ideas, 21 00:00:49,140 --> 00:00:51,660 and these will probably be most of your prompts. 22 00:00:51,660 --> 00:00:53,520 Generally when using ChatGPT, 23 00:00:53,520 --> 00:00:54,990 you just have a quick question 24 00:00:54,990 --> 00:00:57,240 and you don't really need to go into detail 25 00:00:57,240 --> 00:00:59,730 with crafting an entire prompt. 26 00:00:59,730 --> 00:01:01,620 But on the other hand, when you do need to do that, 27 00:01:01,620 --> 00:01:04,590 the advanced section is gonna go into the intricacies 28 00:01:04,590 --> 00:01:08,100 of creating detailed outputs that you can use for personal 29 00:01:08,100 --> 00:01:09,990 and professional purposes. 30 00:01:09,990 --> 00:01:12,360 So what is prompt engineering? 31 00:01:12,360 --> 00:01:15,360 Prompt engineering is the process of designing 32 00:01:15,360 --> 00:01:20,360 and optimizing your prompts using NLP models like ChatGPT 33 00:01:20,610 --> 00:01:23,520 and this involves crafting the prompt. like we mentioned, 34 00:01:23,520 --> 00:01:25,770 in a clear, concise, and effective way 35 00:01:25,770 --> 00:01:27,870 so you can get your desired output. 36 00:01:27,870 --> 00:01:29,730 And I put a little bit of a metaphor here. 37 00:01:29,730 --> 00:01:31,230 AKA, prompt engineering 38 00:01:31,230 --> 00:01:33,990 is like making an effective fishing lure. 39 00:01:33,990 --> 00:01:36,930 Just like a well-designed lure is more likely to go 40 00:01:36,930 --> 00:01:39,360 and catch the fish you want when you're fishing, 41 00:01:39,360 --> 00:01:41,430 a well-crafted prompt is more likely 42 00:01:41,430 --> 00:01:45,690 to get your specific response or desired output. 43 00:01:45,690 --> 00:01:48,660 And now onto the three primary principles 44 00:01:48,660 --> 00:01:50,070 of prompt engineering. 45 00:01:50,070 --> 00:01:51,870 First up, you wanna be specific. 46 00:01:51,870 --> 00:01:53,310 The more criteria you give, 47 00:01:53,310 --> 00:01:55,380 the more focused the output will be. 48 00:01:55,380 --> 00:01:57,120 You also wanna work in steps 49 00:01:57,120 --> 00:02:00,540 and break up your tasks into smaller chunks, 50 00:02:00,540 --> 00:02:02,070 just as you would with a human. 51 00:02:02,070 --> 00:02:03,150 You wanna go to your friend 52 00:02:03,150 --> 00:02:05,310 and ask them to do 10 things back to back. 53 00:02:05,310 --> 00:02:07,470 You would work through things methodically. 54 00:02:07,470 --> 00:02:09,720 First, let's create our schedule for the week, 55 00:02:09,720 --> 00:02:13,170 then let's talk about our schedule for the next month, 56 00:02:13,170 --> 00:02:14,880 then let's talk about our long-term goals, 57 00:02:14,880 --> 00:02:17,040 and then let's go here, here, and here. 58 00:02:17,040 --> 00:02:19,470 So with ChatGPT, you wanna do the same thing 59 00:02:19,470 --> 00:02:21,270 and break larger tasks. 60 00:02:21,270 --> 00:02:23,670 Don't ask it to write an entire book. 61 00:02:23,670 --> 00:02:26,880 Start with the chapters or start with the table of contents. 62 00:02:26,880 --> 00:02:30,300 Go into the chapters, go into the content of each chapter, 63 00:02:30,300 --> 00:02:32,340 and then further, further go into the details. 64 00:02:32,340 --> 00:02:34,920 This is gonna make a better output. 65 00:02:34,920 --> 00:02:37,020 And lastly, on these three principles, 66 00:02:37,020 --> 00:02:38,850 is you wanna iterate and improve. 67 00:02:38,850 --> 00:02:41,100 Like we mentioned, one of the most amazing parts 68 00:02:41,100 --> 00:02:43,920 is that you can rework your inputs 69 00:02:43,920 --> 00:02:48,920 to have ChatGPT iteratively improve on its own output. 70 00:02:49,200 --> 00:02:51,960 Now let's talk about what makes a good prompt. 71 00:02:51,960 --> 00:02:54,630 So a great output all comes down 72 00:02:54,630 --> 00:02:56,610 to the data that the model was trained on, 73 00:02:56,610 --> 00:02:58,680 like I mentioned, its parameters, 74 00:02:58,680 --> 00:03:01,890 AKA how it's trained to use the data, 75 00:03:01,890 --> 00:03:04,860 and then your input, AKA prompting. 76 00:03:04,860 --> 00:03:06,690 So since we can only control one of these, 77 00:03:06,690 --> 00:03:09,150 we can't really control the model that it was trained on 78 00:03:09,150 --> 00:03:10,680 and we can't control the way 79 00:03:10,680 --> 00:03:12,570 it was programmed to use that data. 80 00:03:12,570 --> 00:03:14,880 We can only control our prompting. 81 00:03:14,880 --> 00:03:16,680 So here's what that boils down to. 82 00:03:16,680 --> 00:03:19,980 Clear and concise language that's direct and unambiguous. 83 00:03:19,980 --> 00:03:24,450 The persona that you wanna assign ChatGPT to act as. 84 00:03:24,450 --> 00:03:28,170 For example, you can act as a marketing expert, 85 00:03:28,170 --> 00:03:32,550 a biologist, a mathematician, or a psychologist. 86 00:03:32,550 --> 00:03:34,590 And then it's about the information 87 00:03:34,590 --> 00:03:36,450 and examples that you provide. 88 00:03:36,450 --> 00:03:38,400 And the interesting part about this 89 00:03:38,400 --> 00:03:41,970 is that ChatGPT is going to basically remember 90 00:03:41,970 --> 00:03:44,850 and be able to reference any of the information 91 00:03:44,850 --> 00:03:48,030 and examples in the previous chat history. 92 00:03:48,030 --> 00:03:50,280 So even if it's not in the direct prompt 93 00:03:50,280 --> 00:03:52,710 that you're inputting in that very moment, 94 00:03:52,710 --> 00:03:54,360 if you were having a conversation 95 00:03:54,360 --> 00:03:57,600 and you pasted example information into the chat 96 00:03:57,600 --> 00:03:59,430 earlier in that chat history, 97 00:03:59,430 --> 00:04:01,410 it's gonna be able to reference that data. 98 00:04:01,410 --> 00:04:03,600 Next is a specific task 99 00:04:03,600 --> 00:04:07,410 that you are clearly requesting ChatGPT to complete, 100 00:04:07,410 --> 00:04:10,080 AKA your desired output. 101 00:04:10,080 --> 00:04:13,890 And lastly on what makes a good prompt is refinement. 102 00:04:13,890 --> 00:04:16,500 So once you've received your first response, 103 00:04:16,500 --> 00:04:17,820 you can then reiterate 104 00:04:17,820 --> 00:04:20,670 until you receive your desired output. 105 00:04:20,670 --> 00:04:22,860 So that's an overview of what makes a good prompt. 106 00:04:22,860 --> 00:04:25,920 Now let's actually talk about the main prompting steps 107 00:04:25,920 --> 00:04:27,780 that you wanna keep in mind. 108 00:04:27,780 --> 00:04:31,230 So you first wanna define the goal or the problem, 109 00:04:31,230 --> 00:04:33,690 and you wanna clearly articulate what you want 110 00:04:33,690 --> 00:04:36,420 ChatGPT to help you with, kinda like what we're mentioning 111 00:04:36,420 --> 00:04:37,740 at what makes a good prompt, 112 00:04:37,740 --> 00:04:40,290 this is another way to frame it in your mind 113 00:04:40,290 --> 00:04:41,520 so you're really clear on it. 114 00:04:41,520 --> 00:04:43,800 So you wanna tell it what your goal is 115 00:04:43,800 --> 00:04:46,020 or what your problem is, what you want fixed, 116 00:04:46,020 --> 00:04:47,010 what you want answered, 117 00:04:47,010 --> 00:04:48,840 and you wanna use relevant keywords 118 00:04:48,840 --> 00:04:51,930 and phrases that you can input, 119 00:04:51,930 --> 00:04:55,620 which will help guide things to your desired response. 120 00:04:55,620 --> 00:04:58,140 The next step, of course, is really crafting 121 00:04:58,140 --> 00:05:01,080 and writing your prompt clearly, like we mentioned, 122 00:05:01,080 --> 00:05:03,060 the information, the task that's required, 123 00:05:03,060 --> 00:05:06,060 using keywords and combining everything in step one 124 00:05:06,060 --> 00:05:08,220 in your specified prompt. 125 00:05:08,220 --> 00:05:12,000 Then lastly, you wanna test, evaluate, and iterate. 126 00:05:12,000 --> 00:05:15,270 So we generate our responses, we evaluate the results, 127 00:05:15,270 --> 00:05:17,340 we adjust the prompt as needed, 128 00:05:17,340 --> 00:05:19,350 and then we repeat the process 129 00:05:19,350 --> 00:05:23,250 until we have a successful rockstar prompt that is created 130 00:05:23,250 --> 00:05:26,433 and consistently generates the desired response.