Learning Library

← Back to Library

No‑BS Guide to Effective AI Prompting

Key Points

  • The presenter highlights a widespread gap: most AI tutorials are generic, leaving users with specific, real‑world questions (e.g., comparing financial reports, verifying AI answers, polishing emails) that aren’t adequately addressed.
  • The session promises a hands‑on, example‑driven “no‑BS” AI class that walks learners through concrete prompts, explains why they succeed, and supplies detailed write‑ups for future reference.
  • A key teaching point is prompt engineering: a minimal one‑sentence prompt yields a cold, overly formal response, while adding relevant context and structure produces a far more useful, natural‑tone output.
  • By demonstrating the same deadline‑change email request with a richer prompt—including background, desired tone, and specific constraints—the presenter shows how extra “meat on the bone” dramatically improves the AI’s relevance and actionability.
  • The overall takeaway is that effective AI use hinges on supplying clear, context‑rich prompts, and the presenter aims to equip participants with the exact phrasing and examples needed to get reliable, practical results.

Sections

Full Transcript

# No‑BS Guide to Effective AI Prompting **Source:** [https://www.youtube.com/watch?v=esqPTMDvw7w](https://www.youtube.com/watch?v=esqPTMDvw7w) **Duration:** 00:24:01 ## Summary - The presenter highlights a widespread gap: most AI tutorials are generic, leaving users with specific, real‑world questions (e.g., comparing financial reports, verifying AI answers, polishing emails) that aren’t adequately addressed. - The session promises a hands‑on, example‑driven “no‑BS” AI class that walks learners through concrete prompts, explains why they succeed, and supplies detailed write‑ups for future reference. - A key teaching point is prompt engineering: a minimal one‑sentence prompt yields a cold, overly formal response, while adding relevant context and structure produces a far more useful, natural‑tone output. - By demonstrating the same deadline‑change email request with a richer prompt—including background, desired tone, and specific constraints—the presenter shows how extra “meat on the bone” dramatically improves the AI’s relevance and actionability. - The overall takeaway is that effective AI use hinges on supplying clear, context‑rich prompts, and the presenter aims to equip participants with the exact phrasing and examples needed to get reliable, practical results. ## Sections - [00:00:00](https://www.youtube.com/watch?v=esqPTMDvw7w&t=0s) **No‑BS Practical AI Guide** - The speaker pledges a hands‑on session that addresses real‑world AI doubts—validating answers, refining rewritten text, comparing financial reports—by demonstrating concrete prompt examples and clear, actionable explanations. - [00:03:13](https://www.youtube.com/watch?v=esqPTMDvw7w&t=193s) **Implementing Shared Deadline Tracker** - The speaker proposes using a lightweight shared tracker and weekly snapshots to log deadline changes, allowing AI-generated nudges and process refinements for more effective real‑world task management. - [00:06:29](https://www.youtube.com/watch?v=esqPTMDvw7w&t=389s) **Prompt‑Guided Meeting Note Extraction** - Demonstrates how a precise prompt directs ChatGPT to generate concise, risk‑focused meeting summaries—listing key decisions, open questions, and upcoming actions—while strictly avoiding generic fluff. - [00:11:14](https://www.youtube.com/watch?v=esqPTMDvw7w&t=674s) **Personalized Prompt Engineering for Finance** - It explains the need to train users to craft tailored prompts—such as driver‑delta or factor‑attribution queries—for financial analysis, avoiding generic, bulky prompt collections. - [00:14:24](https://www.youtube.com/watch?v=esqPTMDvw7w&t=864s) **Explaining Structured Prompting Simply** - A parent requests a child‑level explanation of why a typo‑filled prompt succeeds, illustrating how clear, structured prompts guide the AI to provide relevant, helpful responses. - [00:18:04](https://www.youtube.com/watch?v=esqPTMDvw7w&t=1084s) **Iterative AI Email Refinement** - The speaker walks through step-by-step prompts to improve an AI‑drafted email, emphasizing clear subject lines, removal of buzzwords, added warmth, and placeholder details. - [00:22:08](https://www.youtube.com/watch?v=esqPTMDvw7w&t=1328s) **Effective Prompt Strategies Explained** - The speaker explains why certain prompts reliably work, demonstrates specific and catch‑all examples, and encourages applying them to various job tasks. ## Full Transcript
0:00This is the no BS guide to actually 0:03using AI. I get so many questions that 0:06are actually really reasonable and it 0:07tells me that AI is doing a disservice 0:10to all of you. Like the AI can't teach 0:12it. Google can't teach it. Most of the 0:14people out there are spouting generic 0:16nonsense and you're left asking me 0:18really specific questions like Nate, how 0:21do I compare two financial reports? How 0:24do I know which one is right? or Nate, 0:26AI is giving me an answer and I don't 0:28know if I should believe it or not. Or 0:30Nate, the rewritten email doesn't sound 0:32right. How do I make it sound better? 0:34Over and over again, people are asking 0:36these questions. They're right to ask 0:38them. AI is supposed to be able to help 0:40with this stuff, but we in the teaching 0:43community are doing you all a disservice 0:45because we haven't communicated it 0:47clearly enough with enough good 0:49examples. That's what I'm here to do 0:51today. That is what this time is for. We 0:53are going to get on screen. We're going 0:55to look at some specific examples. We're 0:57going to explain why they work. And 0:59there's going to be lots more examples 1:01in the write up. I want you to walk away 1:03feeling like this is the most concrete, 1:06specific AI class that you have ever 1:10been to. Like you got to spend a lot of 1:12time with Nate hanging out talking about 1:15prompts and getting your real questions 1:17answered. So, with that in mind, let's 1:19get to it. All right. Our first example 1:22is a real example I got questions about. 1:25Help me tell my manager to stop changing 1:28my deadlines. Well, this is the first 1:30version. This is what a lot of people 1:31start with, just one sentence. And to be 1:33honest, the answer isn't too bad, right? 1:36Uh the AI comes back. It thinks for 17 1:39seconds and it gives you an email that 1:42you can send. It gives you a live script 1:44you can deliver. It feels a little bit 1:46cold. Like who wants to tell their 1:48manager, I want to deliver predictably. 1:50When deadlines change after we commit, 1:52we incur rework and slip risk. That 1:54sounds like Chad GPT5 talking, not a 1:56person. Uh, and then you have an 1:58escalation ready and it just keeps going 2:00and going and going and what is your 2:02next step and it tries to give you hope. 2:03I will say it gives you a lot for that 2:05one line, but my question is how usable 2:08is it? How usable is it for that one 2:10line? For example, does it actually work 2:12for you to freeze dates 48 hours before 2:14kickoff or did it just make that up? You 2:17get the idea. The more context you give 2:19the AI, the more it's going to be useful 2:22for you. Why don't we see an example 2:24that is exactly the same question, but 2:26with just a little bit more meat on the 2:28bone for the prompt. And let's see how 2:30Chat GPT5 handles this. Okay, the first 2:33thing that we see is that this is a 2:35longer prompt. Let's break it down and 2:36see what we did. Is it scary? Not too 2:39scary. Uh, it's got some labels, but my 2:41manager changes project deadlines every 2:43week. Usually mentions it in passing. 2:45That's already useful. This is what I 2:47need you to do. Set up a simple process 2:49for deadline changes without seeming 2:50difficult or rigid already. This is 2:52differing from the answer chat GPT5 gave 2:55previously. These are my these are my 2:57boundaries. I can't send sound 2:58accusatory. I can't damage the 3:00relationship. This is what success feels 3:02like to me. And then this is the 3:05audience, my manager. This is the tone, 3:06etc. So, this is very specific, right? 3:09Uh here's what you get. You get a very 3:13specific set of options. You either get 3:16the lightweight tracking option. I 3:19really value how quickly you move things 3:20forward. Your speed helps us respond to 3:22shifting priorities is very much the 3:23sandwich approach to conflict, right? 3:25But when deadlines shift in 3:26conversations, I can miss the update and 3:28realize later. What do you think about 3:30us logging deadline changes in a single 3:32shared tracker? This is a talk track I 3:35could see actually using. This is 3:36immediately more useful. If you want to 3:38go farther, you actually have the option 3:41to write down a little nudge and use 3:45that as part of the method. And so this 3:46is where Chad GPT is actually suggesting 3:49what if whenever you shift a date, I jot 3:50it down and send a oneliner back. This 3:52is both a script and also a suggestion 3:55for a change in process in one go here. 3:57And then there's also the weekly 3:58snapshot. What if we do a weekly 4:00snapshot where we look at the deadlines 4:02and make sure it's right and then it 4:03comes back with if they push back, how 4:05you handle it, assumptions about their 4:08motivation. So it logs it and then you 4:10can refine it from there. What do you 4:12see here? What I see if you compare it 4:14to the initial version is that this is 4:17actually usable for a realworld 4:20situation. And people who are getting 4:22started with AI will say, well, how did 4:24you get it to do that? What is the magic 4:26in this prompt? Well, to be honest, a 4:29lot of the magic in the prompt is giving 4:32the AI enough about you to be helpful. 4:36So, this detail changes project 4:38deadlines weekly. You see that coming 4:40back up in the options. This detail set 4:43up a simple process. You see each of the 4:44options has a simple process. If you 4:47look through this prompt, you can see 4:50that chat GPT has taken each of the 4:53words you've given it really seriously 4:55and tried to come up with something that 4:57puts it all together. And so what you're 4:59giving the AI in a prompt is not a set 5:02of magic words. It's actually enough 5:04context, enough um information about 5:07what you're doing that the AI can be 5:09helpful. So my manager values 5:11flexibility and speed. You'll notice 5:13these talk tracks mention that back 5:15because that's considered good practice. 5:18This is not magic. This is the AI coming 5:20back and trying to mirror what you give 5:22it. If you give it more, it will give 5:25you back more. Let's check out another 5:27real life example with a real life 5:29question mark. I love this one because 5:31it looks like a long prompt, but it's 5:33actually a very realistic short prompt. 5:36This is the prompt up here. Summarize my 5:38meeting notes. That's it. And then this 5:40just pastes in a bunch of meeting notes. 5:42In this case, to protect people's 5:43privacy, I actually had another AI make 5:46up the meeting notes. So, anyway, here 5:47are the meeting notes. There's a 5:48complete transcript. Uh, and then it 5:51comes back with meeting notes, right? 5:52And again, I don't want to underell it. 5:54There's some value here. It captures the 5:56outcomes in the notes. It captures the 5:58feedback. It captures some action items. 6:01This is not too bad, but it's really, 6:03really easy to do a little bit better 6:06than that and make these notes much more 6:08useful. In particular, if you have ever 6:11been in meetings, you know that one of 6:13the really scary things about meetings 6:15is the idea that when we talk back and 6:17forth, sometimes we slide over project 6:20risk that needs to be written down, 6:22formalized, and talked about or else it 6:24just kind of slides under the radar and 6:26we end up discovering it on launch day. 6:28If you've been a project manager, a 6:29product manager, an engineer, uh, in 6:32marketing, you know that feeling. It's 6:34that clenching in your gut. How do you 6:36avoid that? These meeting notes don't 6:38help with that, but that's one of the 6:39most important things that you could 6:41actually get. These risks are fairly 6:43generic. GA delay strains marketing 6:46timelines. Okay, that's fine. Like, 6:47that's really generic. Let's see how a 6:50better prompt with the same transcript 6:52could change things. Okay, the first 6:54thing we notice is there doesn't seem to 6:55be a prompt here, but don't worry, there 6:57is. This is the meeting transcript. If 6:59we click show full message here and we 7:01scroll down, we eventually get to the 7:04prompt right here at the bottom. pull 7:06out and organize. I want a summary of 7:08the main things that happened. I want 7:09key decisions. I want open questions. I 7:12don't think that that one was very 7:13clearly pulled out before. I want risks. 7:15I want the next seven days. And I and 7:17you define the risks. You see how you 7:19define the risk with likelihood impact 7:21and who's on it. And please skip a 7:23section with nothing in it. Please don't 7:25add fluff. That's critical. And you'll 7:27notice how obsessed chat GPT is with 7:30following this. It immediately comes 7:31back with summary of five main things 7:33that happened. It is trying to tell you 7:35with this word that it has paid 7:37attention to don't add fluff or context 7:39and it's taking it really seriously. It 7:41identifies the onboarding bottleneck, 7:43the performance fix, the GA launch, the 7:45pricing tiers. It it gives you all of 7:46this stuff. Okay, so these are your top 7:48five. It gives you decisions that 7:51happened. It gives you open questions 7:53and it gives you risks that are much 7:55much more interesting. 7:57And this by itself is worth the price of 8:00admission for the prompt because you can 8:02then look at that meeting and you can 8:04say, "Oh, you're right. Marcus and David 8:08are supposed to be watching the 8:09performance fixes, but I'm not clear 8:12that they actually have it after I kind 8:14of think through the meeting. Maybe I 8:16should follow up with them." This pulls 8:18out those things that are intangible in 8:20the meeting and makes them tangible. In 8:22other words, chat GPT is actually adding 8:25value here. And it's not adding value 8:27because it's magic. It's adding value 8:29because it is following the instructions 8:31that you gave it. If we come back here, 8:33all we had to do to get that kind of 8:35magical response was to say, "Please 8:38tell me what could go wrong very 8:40specifically, and then say this is what 8:41I care about. I care about the risk, how 8:43likely it is to happen, the impact, and 8:45who's watching it." These are not magic 8:47words. If you wrote uh the risk, the 8:50probability that it's going to occur, uh 8:52what will happen if it occurs, and the 8:54owner, it would still mostly just work 8:56the same way. The point is, can you 8:59explain in really clear words what you 9:02want? Can you explain what matters to 9:04you? And I think we're hanging out here, 9:07right? This is the chat GPT classroom. 9:10One of the things that people really 9:12miss with AI that they struggle with 9:14with AI 9:16is this idea that you can ask for 9:19something that you want that you haven't 9:21seen before done well and it can still 9:24work. When I was keeping notes manually 9:26and yes I did that for a long long time. 9:28I did not often see risks as clearly 9:31laid out as I just showed you. That was 9:34rare. But now, even if I don't have a 9:36good example beyond saying this is what 9:38I want, I can still get it done. And so, 9:42one of the things that I like to tell 9:44people in Chad GPT class is 9:48have the courage to imagine a better way 9:51to do things and then find the words to 9:54express it. That is one of the biggest 9:56keys to prompting. And I find that 9:58people have these light bulb moments 10:00when that happens. are like, "Oh, I can 10:03just ask for it." Yeah, you can ask for 10:05it. You can ask for it. Let's look at 10:08another example, but we're going to make 10:09it more fun. I love this example because 10:12it's a what do I do when I don't have a 10:14prompt example. I have a situation not 10:16covered in Nate's guide. Right? Here's 10:18the situation. I need to compare two 10:20financial reports. I need to understand 10:22the drivers for financial performance 10:24between them. I need you to give me an 10:26answer in 100 words or less. And the 10:28audience is financial professional. Now, 10:31you might think that's the prompt. That 10:33is the context. And I'm going to explain 10:35how you explain that to the AI. It's 10:37very clear. Here's my context. It's four 10:39different bullets right now. This is 10:41what the you're actually asking the AI 10:43to do using the guides methods. Uh and 10:47then you define them here. Ask do clear 10:50format delta only revisions, which means 10:53revisions that are only about the 10:54difference between situations and 10:56verification. Please make a brief 10:59outline of two approaches to handle this 11:01situation. A full prompt to paste into 11:03AI, including the context, the 11:05instructions, the format, and the 11:07verification step. And then please 11:09explain why this works. 11:12That is what you're asking AI to do. 11:14This is an example of asking chat GPT to 11:19write the prompt for you. Your real 11:21instruction here is write the prompt for 11:24me. And if you were in the business of 11:26teaching AI, you have to be able to 11:28explain this because one of the things 11:30that we learned from that survey of 700 11:33million uh chat GPT users that came out 11:35this week is that we have such a wide 11:38range of use cases for AI that we need 11:41that personalized perspective. You need 11:43to be able to teach people to prompt 11:46like this so that they don't have to 11:48carry around a gigantic sheet of all of 11:50these different prompts. like they need 11:52something that works for more than one 11:54thing. So then chat GPT comes back and 11:56it comes back with two options. You have 12:00what it calls a driver delta. It can 12:02highlight changes in revenue, cost, 12:04margins or cash flows between reports A 12:06and B or it can come back with uh 12:10attributing performance shift to 12:12particular volume, price mix or cost 12:14controls and quantify the impact. You 12:17can choose. And so for example, so 12:19here's the full prompt. It's assuming 12:21key drivers. It's assuming number one 12:23here. Uh 12:26we can say actually 12:30please write the factor attribution 12:34approach 12:36prompt instead and it will come back 12:38with the other prompt. And there you go. 12:41It's a full prompt. It's ready for you 12:43to attach those files and get going. Not 12:45by the way because there's a magic word 12:48up here. The only thing you did up here 12:50was you took what would be a generic 12:52prompt and you made it into something 12:56financial with the words that you used. 12:58This is what is telling Chad GPT it's a 13:00financial prompt. You can put anything 13:02in there. You can change it entirely. As 13:05an example of changing entirely, let's 13:07just do this real quick. We'll do it 13:08live. Okay, here we are. We're going to 13:10delete all of the financial report 13:12stuff. You're going to watch this run 13:14live. Going to delete all of these 13:15things. This is the only thing we have 13:16to delete, right? We deleted all of this 13:18stuff. That's super easy. We're going to 13:21change the situation. I need to talk to 13:24my kids teacher about her homework. My 13:28goal have a constructive conversation 13:31and get strategies to help with reading 13:35chapter books. 13:38Constraints 13:39don't want to 13:42break the relationship with the teacher. 13:44I'm deliberately leaving that typo in. 13:46Actually, I want you to remember you can 13:47leave typos in audience. Uh the teacher. 13:50Um 13:52and so I need a prompt for this, right? 13:54I don't know how to even approach this. 13:56Um please help me figure it out. So I 13:58hit go. So it gives me different 14:00options. You can say frame is wanting to 14:02work together. Share a neutral 14:03observation. Um and then it gives me a 14:06frankly nice prompt. I need to talk to 14:08my child's teacher. Please generate two 14:10conversational scripts using a 14:12collaborative ask and an observation 14:14ask. revise into one recommended 14:16version. I love this. I never came up 14:18with this. Chad JPT did this. And then 14:20it explains to me why it works. 14:22Um, 14:24please run the prompt. And then, and you 14:27notice it worked with a typo. And then, 14:29please explain to me like I'm 12 why it 14:33works. Okay. This is giving me words I 14:36wouldn't have been able to find. I 14:37really appreciate how much effort you've 14:38put into helping my child build her 14:39reading skills, especially if it's an 14:40emotional thing. As a parent, these 14:41kinds of scripts are helpful. gives me a 14:43recommendation. Uh, it gives me a sense 14:46of why it thinks it works and it gives 14:49me a sense of the why. Right? If you if 14:51you say you're the expert, what should I 14:52practice? The coach feels respected. It 14:54gives me a metaphor with soccer that 14:56works. Uh, I'm going to go one step 14:57further here to close out our example. 15:00Uh, can you explain uh like I'm 12, 15:04again, I'm leaving the typo, why this 15:06prompting approach works well. Um, how 15:09does prompting like this help you as the 15:13AI help me better? I love this. Think of 15:16me as a fast, eager helper. I sometimes 15:18get carried away. If you say, "Help me 15:20talk to my kids teacher," I might give 15:22you something polite but random because 15:23I don't know what's most important. But 15:25when you use structured prompting, 15:27you're telling me what question you want 15:29me to answer, what action you want me to 15:30take. You're giving me constraints, and 15:32that ultimately helps me help you. 15:34Fantastic. Yes, chat. GPT can help you 15:37prompt. Let's look at one more example. 15:39This one is almost never covered in 15:41classes, and I think that's 15:42disappointing. Most of the time, people 15:44will give you a single prompt like I've 15:46been showing you, and then they'll be 15:48like, "Go and be well. You have your 15:50prompt." You know, you have Nate's 15:52master prompt. You're off you go. Well, 15:55what if you aren't that person? What if 15:58you're having a conversation? How do you 15:59have a conversation that progressively 16:01reveals the information you need? So, 16:03for example, again, a real question. You 16:06get to a more human sounding email. And 16:08so, here's an initial situation. 16:09Already, this is fine, right? I run a 16:11design agency. A client's been with us 16:13for 3 years. They want to rush the 16:14project. I need to turn them down nicely 16:16without ruining the relationship. Please 16:19keep it short. Uh, please don't use bud 16:21buzzwords. Uh, and if they can 16:25understand that that we can't help now, 16:26but they'll reach out later, that's 16:28great. It comes back. This is okay. 16:32It's kind of it still feels wordy. I 16:35feel like I would not send this because 16:37it feels like a wall of text to a 16:39client. Uh, and so I want to give the I 16:41want to give the the AI a little bit 16:43more context to help me. Okay, so this 16:45is a CMO. I worked with them for years. 16:46You need you need to know that, right? I 16:48need friendly but professional tone like 16:51I'm talking to a colleague. Um, and 16:54someone else they could work with is 16:56something that I want to include here 16:57because you just realize that right as 16:59you're reading the draft. So comes back. 17:00Uh, it's a little better. I'm grateful 17:02you reached out. We've really valued 17:04working together. There is a 17:05recommendation in here. Uh, it still 17:07feels a bit like a wall of text. Um, and 17:11so I just want some advice. And so I I 17:13just ask that, right? Actually, you know 17:15what? Before writing anything, please 17:16give me two different ways I could 17:18handle the decline. Ask me a question 17:20about the situation that might help you 17:21to write better. And this is a great 17:23example in a multi-turn conversation 17:25where you can actually throw the ball 17:26back to chat GPT and you can ask it to 17:28ask you, right? you can ask it to give 17:31you tips and that will be really 17:34helpful. So it comes back, it says you 17:35can protect the relationship with a 17:37really clear referral or you can 17:38position it very much as a timing issue 17:40and emphasize future work more. And so 17:42it's taking from what you've told it and 17:44saying which do you want me to emphasize 17:46and then it's asking do you want the 17:48referral to be the heart of the message 17:50or something that's lighter that you 17:51might try in the meantime and that's a 17:53really nuanced question that we haven't 17:55given a perspective on. And so I looked 17:57at that and I'm like, I think I want a 17:59clear referral. You could have picked 18:01either one, but I went with option one. 18:02All I say is go with one. It drafts it 18:04out. This is already better. It's two 18:06paragraphs. It's more scannable rather 18:09than risk a rushed outcome. It's still a 18:10little cold. Uh and I feel like I'm 18:13worried that the buzzwords are still in 18:15there. So I'm saying skip the buzzwords. 18:17I need to have the subject line really 18:19clear and I need to have the email 18:21itself. And I that's that's all I want 18:23us to sort of focus on. I don't want it 18:26to be I don't want it to be anything but 18:28what I need to check and read. And I 18:29feel like right now when I look at it, 18:31it's better, but there's still something 18:33missing. And so I'm basically trying to 18:34get the AI to give me some clarity, 18:36especially around the subject line. So 18:38now we have this a quick note on the 18:40Rush project. Uh on timing for this 18:43project as a backup. Uh this is already 18:46better. It dumps the buzz. Dumping the 18:48buzzwords was really good, right? I'm 18:49glad you thought of us. We've loved 18:50working with you. That feels really 18:52natural. the M dash might be a little 18:54bit of a giveaway depending on who you 18:56believe about chat GPT writing but still 18:58it's it's not bad uh and then I ask it 19:02this look at what you wrote and I want 19:04you to tell me is it specific enough are 19:05we naming dates and steps now I haven't 19:07given it dates and steps but I want 19:09placeholders does it sound warm enough 19:11please list five things you changed to 19:13improve it you see how I'm basically 19:15leading the AI through an edit 19:17conversation I'm not just leaving it 19:19alone I'm not saying this is good enough 19:22people might have taken these and said, 19:24"Ah, I don't know. I don't know. I don't 19:25know." Uh, and you might be saying to 19:27yourself, "This is a lot of edits for an 19:29email." But think about it. This is a 19:30client that may be a six-f figureure 19:33client for this for this small firm. 19:36They don't want to lose them. They just 19:37need to delay things a bit. It 19:39absolutely is worth a 5minute 19:41conversation with Chad GPT to get to a 19:43better email. Okay. Is it specific 19:45enough? Chad GPT admits not quite. Does 19:48it sound warm enough? It's clipped. I 19:49agree. Here's some things I could do to 19:52include it or in not include it, improve 19:54it. Here's a revised version. We valued 19:56your trust for this rush request. I need 19:58to be honest. We can't give it the focus 20:00it deserves. I recommend this is very 20:02readable. It looks like it's almost 20:04there. Uh for this final version, tell 20:07me what assumptions are you making? How 20:08confident are you that this keeps the 20:10relationship strong? And what's one 20:11thing I should double check? You can 20:13actually ask them to double check it. 20:15Make sure the referral you name is 20:17reliable as the double check 20:18recommendation. I agree that sounds like 20:21an important. So there you go. What we 20:23have here is something that is now ready 20:25to paste in that is quite strong. It may 20:27not be perfect but it is strong enough 20:30that it is 90 95% there for a delicate 20:34email to a highv value six or seven 20:36figure client. We did not start there. 20:38You do not have to start with perfect 20:41with AI to get it where you want to go. 20:43And I want to go back through and note 20:45we're sort of following stream of 20:47thought and all we're doing is taking 20:49enough time to add the words so the AI 20:52knows how to key off of us so that it 20:54can be as helpful as possible. List 20:57exactly five small things you change to 20:58improve it is just a little bit more 21:00precise than tell me what you'd fix. 21:02Those are the kinds of small differences 21:04that aren't magic. They just help the AI 21:06help us. I hope that this has been a uh 21:09helpful multi-turn conversation to look 21:12through. I know we don't do a ton of 21:14these and I think it's really important 21:15to look at them because they show how we 21:17can follow our own train of thought and 21:19still get somewhere really useful with 21:21an AI. So, I want to close by going back 21:24to some of the things that people have 21:25been asking me initially and reframing 21:28them as larger questions that I see all 21:29over the internet that I see in my inbox 21:31and how we're answering them. So, how do 21:34I get better responses out of AI? That's 21:36one of the underlying questions. All I'm 21:38saying is give the AI enough context to 21:42help you. When I gave you those answers, 21:44I on purpose used the same kinds of 21:49labels in every single one. The 21:51situation, what I need, my constraints, 21:54what success looks like. You can name 21:56those different things. But those are 21:57some basic categories of meaning, 22:00categories of words that if you give 22:03that to AI, it can go farther for you. 22:06What prompts actually work? That's 22:08another one that I get a lot. These 22:11prompts work. I gave you real scenarios. 22:13They actually work. And they don't work 22:15because I wrote them. They don't work 22:17because they're magic. They work. And I 22:19hope you can see this because they give 22:22the AI the information it needs. I also 22:25hear a lot, I need examples for my 22:27specific job. Whether that's boss 22:29conversations, emails, meeting notes, 22:31whatever. I gave you some of those 22:32specific examples, but inevitably you're 22:34going to think of some that I didn't 22:36cover because this is only going to be 22:37so long. That is what those larger 22:39catchall prompts are for. And that's why 22:41I deliberately illustrated one of my 22:44examples, the financial analysis one 22:46with a catch-all prompt. You are 22:48empowered to go take that kind of a 22:50prompt and put it to use for your 22:52situation. And that is why I spent so 22:54much time showing you the actual 22:56prompts. It's not because I wanted to 22:57bore you. It's because I wanted you to 22:59understand how they work and I wanted 23:01you to see them really play out side by 23:03side so you could see I'm not just 23:05trying to make stuff up. I'm not just 23:07trying to tell you more words are 23:09better. I'm trying to give you clarity. 23:10I'm trying to give you the ability to 23:12get help from the AI. So if you take 23:14anything away from this class on AI, 23:17please take away a sense that it is not 23:21that hard to get to can use AI in many 23:25of my tasks every day. The prompts that 23:27I have just laid out for you covers so 23:30many different real work situations. All 23:32you have to do is slightly change the 23:34situation and what you need from the AI 23:37and your goal in situation and the 23:39context you give it and you're going to 23:41be off to the races. You can use an AI 23:44to get a better answer. You can shape 23:46what you get from chat GPT so it is more 23:49useful and we've seen so many examples 23:51of how to do that. Over to you. Best of 23:53luck and enjoy chat GPT. It is really an 23:56amazing tool and I hope that this has 23:58helped you use it better.