Mastering Prompt Engineering for Better AI Conversations and Productivity
Episode 3 - Mastering Prompt Engineering for Better AI Conversations and Productivity
===
[00:00:28] Tom Adams: Hey, greetings and salutations, a heartfelt welcome back to this episode of the Prompt and Circumstance Podcast, where we try and make sense of all things AI, code and culture, and find a pathway forwards in all the complexity. Joining me today are two of my three colleagues, Ryan nei. And Mike Richardson, the illustrious Mark Redgrave is celebrating his birthday in some far off land with palm trees and clear blue seas.
So he's not with us today.
[00:01:00] Mike Richardson: guy.
[00:01:01] Tom Adams: I know gentlemen. hello to everybody.
[00:01:04] Mike Richardson: Hey, everybody.
[00:01:06] Tom Adams: Hey, we're glad you're here. So, uh, gentlemen, let's jump in. Um, it's been two weeks since our last confession, and I think it's only appropriate to confess what we've been working on. What's been interesting. What are we seeing? What are we aware of? Uh, Mike, let's start with you this week.
[00:01:22] Mike Richardson: Yeah. Well, you know, as I said last time, I've been, uh, playing with my new Surface Pro copilot, Snapdragon. You name it, I've got it. And I'm enjoying exploring all of that. Plus I've got the NewLaw, memory device that. I can record and, uh, and interrogate at the end of the day. But on Ryan's recommendation, an episode or two ago I picked up the book, the AI Driven Leader, I think it's by Jeff Woods, and I'm starting to listen to that. through Audible as I do my workout and sort of daily routines I'm really enjoying it. I'm a couple of chapters in and, uh, the number one thing that I've taken away so far is what he's really making the case for is that is a strategic thinking. Supplement. It's a thought partner for strategic thinking
[00:02:18] Tom Adams: Hmm.
[00:02:18] Mike Richardson: I really, and, and decision making and I really like the clarity of that.
So that's what I've been up to, Tom.
[00:02:25] Tom Adams: Very cool, Ryan.
[00:02:26] Ryan Neimann: Mike.
[00:02:27] Tom Adams: Yeah. Yeah.
[00:02:28] Ryan Neimann: Fantastic.
[00:02:29] Tom Adams: Yeah. And what about you, Ryan? What have you been, uh, working on?
[00:02:33] Ryan Neimann: Well, I think some of it'll lead into some of the things I wanna demo and showcase today. I, I really leaned in on AI literacy and how do you enable that AI literacy, much like. Mike is talking about. How do you start to make that transition to having some sort of discourse with AI that can challenge your thoughts and elevate your strategic thinking?
So I think we've got some good stuff today to review.
[00:02:58] Tom Adams: Oh yeah, I'm excited. Well, I'll just give you my, uh, my short form version of what I've been working on. But, uh, I too, Ryan, since our last conversation two weeks ago, uh, have sort of unpacked your prompt structures and things like that. 'Cause I think a number of weeks ago I talked about using voice as my Primary typing tool and I've, I've got this cool, this tool called Super Whispers. And in super whispers, I can create prompts based on what my voice says. And so I've prompted a prompt to create code based prompts to follow your recommendations.
[00:03:33] Ryan Neimann: There
[00:03:34] Tom Adams: And so now I talk and instead of it just sort of cleaning up my script, basically the transcription, it now actually creates a.
A prompt that builds software, does any kind of things based on your structure. So I, I'm still working on it to try and make that, that work better. But, uh, man, that's, that's been interesting. I think we're sort of focused on prompts, but that's been my first one. Second one is I said I would be testing Warp, which is a terminal tool, which is kind of trying to be like.
Claude code. but my goodness, it's better. So I built a software last week with it. So an app which basically is an HR voice recording software for field managers. So anyone out in the field. Who's kind of managing, say, drivers and stuff like that. If they have an issue with them and they don't have time to type, they just voice it. it goes into a repository. AI cleans it up to look like HR language, puts it in a database so that the HR department can manage it after the fact for, for things. Uh, pretty, pretty cool to be able to do that with no knowledge of what I'm doing. So.
[00:04:39] Mike Richardson: I just want to keep reminding you at all times, as you've heard in previous episodes, and Tom does not know how to code.
[00:04:48] Ryan Neimann: Well done. So that was the, if you see something, say something, application
[00:04:53] Tom Adams: Yeah. Yeah. But I, I went, well, I keep hearing from my coaching clients that one of these challenges is they wish their supervisors who are often active, especially in the Service Bigs business, are active. they can't type their, their hr, um, notes very easily. So I just went and went, oh, there's gotta be a way to do this.
And so I created a Voice Note app for that. So using some of the same toolkit, but I did this with Warp this time, so that was fun.
[00:05:19] Ryan Neimann: Yeah. What I really love about that is it's, able to consume that in a way that's easy for the user.
[00:05:25] Tom Adams: Yes.
[00:05:26] Ryan Neimann: driving away and I need a hands-free way of communicating that. And then translate it into a meaningful way for the enterprise. That's pretty cool.
[00:05:37] Tom Adams: Yeah. Yeah. So anyways, that's, uh, that was my week and, uh, continue to be, uh. Continue to be trying to figure this stuff out in a very practical way. So yeah, so that's, that's where we've been. Um, we know that Mark Redgrave has been just basically inputting sand. Um, so that's pretty much all he's been doing.
Probably sand and salt water, um, and good for him on his birthday. But, um, today we wanna get into the meat of it, and I think we discussed this last time. Uh, we did discuss it, but I, I felt like it might be more important to actually take Mr. Prompt himself, the king of prompts, the, the aficionado of prompt structure and, um, rather than just talk about it, uh, show it to us.
And so I, I think we're gonna admit that this might be better viewed on YouTube than it is. Uh, by voice, but even by voice, you'll pick this up. So if you're listening to it and you wanna make sense of it and see it in action, you could pop over to YouTube and see it, it on YouTube. so Ryan, would you share your screen and show us a little bit of some prompt, sort of some baseline, prompt stuff and then elevate that into maybe more sophisticated stuff the way we talked about it last time.
Marker
---
[00:06:52] Ryan Neimann: Yeah, you bet. Tom. Let me give a, a little bit of, of history, maybe first. So, I think one of the first things I recognized. I would say in 2023, much like everybody else coming into that year, you know, you may have tried, uh, an LLM probably, you know, chat GBT I recognized that the better that I was prompting, the better I was getting responses.
But I seem to recall that year there was just a flood of information on how to better prompt and, and, you know, posts were just filling up my, my LinkedIn feed about. How to do that and improve the outcomes, uh, with, with better prompting. And so I started to, to think about that and, and came up with some tips and tricks.
And I think you'll pick up a number of those. And Tom mentions, this might be better on YouTube, but I'll do my best to provide a bit of an, a voiceover for everybody as well. So what I've done is gone back to some of the tools I used, but I, I went at it as if I were. Maybe somebody who's just starting out or has a base subscription, or doesn't even have a subscription.
I should start there. So I, I, I started off with, uh, chat, GPT and another tool that I'll show you, and we'll kind of step through what that looked like. Um, so this is just your, you know, your basic, uh, chat, GPT. I've, I've logged in and really there's no settings there. I haven't even done my first prompt, so we'll use that as a kind of a construct. the next thing I'll introduce is a platform called poe. Uh, I believe it's named after Edgar Allen Poe. So PO e.com. And you can go to po. And again, I started with just a. know, uh, no subscription or anything, just logged in. And what I think PO does is a really good job of elevating a large number of language models that you can try. mean, really an extensive laundry list, almost, uh, a buffet you could say of, of, of different language models.
[00:08:59] Mike Richardson: You are making me hungry.
[00:09:01] Ryan Neimann: Now, Tom, you, you had said that you had some experience using po.
[00:09:05] Tom Adams: Yeah. I actually played with a subscription for a couple of months and I, think it's got really good bones to it. It's a really cool concept. Uh, I just found that, um. It, it kind of ran outta memory very quickly. It ran out of, yeah. So I, I think it's a fabulous tool to, to get a insight into how different, uh, LLMs respond to something.
but I, I think it's a very interesting tool, look at. Yeah, well, I agree with you. It's, uh, it's almost like a Las Vegas buffet 'cause you get an immense amount of appetizer and dessert, but there's definitely somebody serving out the prime rib. And so we'll see that as we kind of go through,
All right.
[00:09:44] Ryan Neimann: but. so let me kinda show you. So as you log into PO there is a number of bots, in different categories and you can just start to see the laundry list.
And it just keeps going
[00:09:57] Mike Richardson: Oh
[00:09:57] Ryan Neimann: from search bots to image bots, video, audio generation, um, app, uh, bots to make applications. It, it just keeps going. Uh, so you can try any one of these, but maybe as Tom mentions, uh, there are some that are, that are paid for. so one of the things I wanted to illustrate was reinforcing at the core is a statistical correlation of what you. Type in a prompt and what it comes back with the prompt response, right? And so one of the things I started with was what I'll call the bloom sentence example. So I'll switch over just to make the, maybe the demo a little bit easier. Um, what I did is, uh, I went into PO and, and I just prompt, I selected the GPT-3 0.5 turbo role.
Like, just like picked A GPT chat, GPT, uh, for an example.
[00:10:52] Tom Adams: Which is an early one, which is a relatively early one from them. Yep.
[00:10:56] Ryan Neimann: picked that 'cause I wanted to start, you know, like a little bit earlier, like what might have been the response that you would get. So I said, complete this sentence, the balloon is, and micro time. What do you think you know would be the next statistical word that comes to mind?
[00:11:17] Tom Adams: I think red like that. That's the thing that, that, I don't know. The red, red balloons. Yeah.
[00:11:23] Ryan Neimann: Yeah. There's, there's, uh, there's a, you know, so that's the next to Cisco, uh, word that would come, uh, come up. Um, there's also some other combinations of words that would be, you know, words like floating, uh, come to mind. Uh, and, and so those are, uh, all related to the way at its core. And this is a very simple example of what is the next word to finish this sentence.
And so if I look down here. 3.5 came back with floating in the sky and I thought, okay, that's pretty close. I was expecting it to actually say red, but in the sky. And then I thought was interesting is now I want to compare that would other large language models complete it. So I thought I'd stay in the GPT family, right? oh, does a really good job. It just has slash compare. And then. At symbol it'll give you the choice. You can cho choose any number of LLMs, so pick the next one. So I thought, okay, I'll stay in GPT went to GPT five chat and its response was, the balloon is floating gently in the air. Yeah. Okay. Well that's interesting.
So it's a little bit more alliteration
[00:12:39] Tom Adams: Yep.
[00:12:40] Ryan Neimann: kind of, that's interesting. So then I thought I'd go over to Grok. I don't use Grok very often, but you know, I thought I really wanna get a sense of what Grok is about. grok four. Well, that was a paid, uh, version, so I didn't have enough
[00:12:56] Tom Adams: All right.
[00:12:56] Ryan Neimann: think it says to answer the request, and I thought, okay, I'm just gonna go on to a different, maybe earlier version that's free and included with my, my non-paid subscription. And so I compared it with Rock three and it said, thinking. I thought, well, that's, that's interesting.
[00:13:13] Mike Richardson: thinking.
[00:13:14] Ryan Neimann: You know. Well, the, the GPT, yeah, so the LLM is, is, I should say, is, is thinking. And so is interesting, but Groc three came back and really started to explain the way it would complete the sentence. So, you know, it said, it reiterated what I had said, which is complete this sentence. It said, well, it's an incomplete sentence. There's possible completions that are descriptive, narrative, humorous, or educational, I need to reconsider this context. And then
[00:13:47] Tom Adams: Fascinating.
[00:13:48] Ryan Neimann: is, and then it just kind of goes on and it's like, okay, well I need to complete this sentence. So a single completion might be sufficient. Like, oh, okay. Boy, it's really thinking. Now it goes on to the final completion idea, or wait a minute, there's a twist. The balloon is about to pop, and it actually goes through its thinking to come up with final answer, which is the balloon is floating high in the sky, carried away by a gentle breeze.
[00:14:17] Mike Richardson: By the way, everybody, you, you should be, you go to YouTube and look at this because while Ryan has been speaking, he's been scrolling and scrolling and scrolling and, and we've seen all of the different angles of response. That, that Gro gave him. And it's really intriguing to see, uh, these different dimensions that it's explored of how to complete that sentence.
That's amazing.
[00:14:43] Tom Adams: Yeah.
[00:14:44] Ryan Neimann: I, I think it is as well. It illustrates a couple things. One, you know, we talked about on a prior episode, you're in control of what you prompt and you're in control of adding some attachments or custom instructions. But really this is on the logic on the back end and how it takes your prompt and thinks about how it's going to respond or programmatically executes how it's gonna respond. so let's keep going. So I thought, all right, let's compare it to some others. And I thought, you know, I don't use quad as much, so I thought I'd go to quad four. And it said, you don't have enough
[00:15:17] Tom Adams: Yep.
[00:15:17] Ryan Neimann: points for that. And I thought, okay, well I'll go back. So I'm gonna go to Quad SA at 3.7. And it came back with, balloon is floating gently in the breeze.
It's bright colors, catching the sunlight. Well, that's, that's quite interesting. It
[00:15:38] Tom Adams: Well, Claude tends to be more writer focused. It's either software focused or or writer focused, and it'll try to do better writing. For some reason. That's, that's its predisposition.
[00:15:50] Ryan Neimann: The next step, despite what Mark gave me, the warning, I shouldn't be talking to a llama. I decided to talk to a llama. So I went to Llama four Maverick and it just said gently in the air. It didn't, didn't just. and gave the, the, the, the triple lips and just floating gently in the air went on to Gemini 2.5 flash. Ooh, it's thinking
[00:16:16] Tom Adams: Hmm.
[00:16:17] Ryan Neimann: And so it starts thinking, it's formulating possible answers and it gives me four. The balloon is red. Finally we got a balloon is red, the balloon is floating. The balloon is full of air. The last one, the balloon is about to pop. Okay, so now I'm intrigued. I wanna keep going. And I went to Mytral Large.
Now Tommy, you'd said, you know a little bit about Myst Large. I was not as familiar with Mytral.
[00:16:47] Tom Adams: Yeah, it's a European, uh, I think outta France and it's, it's just another smaller LLM company that's building this. And, they've got some good, uh, stuff going on, so.
[00:17:00] Ryan Neimann: It came back with. Here are few possible ways to complete the sentence and without exhausting all of them. There's five. We're talking about high in the sky, red and shiny, all the way to filled with helium. Okay, so I continue to be intrigued, to learn more, and I thought deep seek. Alright. I honestly have never used prior to this moment, deep seek,
[00:17:26] Tom Adams: Oh yeah.
[00:17:26] Ryan Neimann: about it, thought it'd be very interesting to try it. And I'd say this is the most interesting one of all of them. So it said thinking and it came back with, The user wants me to complete the sentence to paraphrase it said, first I should consider why they're asking this. What? Full paragraph on this topic. The word balloon itself has dual meetings.
Well, I really hadn't thought about that. the party decoration or the aircraft. So OO, okay. Then it said, I noticed they used is and not was. To present and I was like, okay, wow. It keeps going.
[00:18:12] Mike Richardson: Is
[00:18:12] Ryan Neimann: And then, then it says, wait, I prioritize educational correctness or playful creativity? Both, I think start with the lit literal mental checklist.
So it must include scientific accuracy and common sensory traits or twists or a safety note. I thought really? It's, it's gonna add a safety note. I, I'm reading this, I'm like, man, this is, this is unbelievable how different this is to respond to this question. It keeps going. It finally says, I just overthink this balloon? But tiny
[00:18:58] Tom Adams: It's absolutely beautiful.
[00:19:00] Ryan Neimann: but tiny prompts can hide big needs. Hmm. deep.
[00:19:06] Tom Adams: That is deep man.
[00:19:08] Ryan Neimann: it keeps going. And it said, it gave me 27 different ways categorically on how to finish this. But remember the choking hazard it said up here? think, uh, let's see if I, here we go. Ah, the choking hazard point matters. I mean, I thought, does it really, I just wanted to finish the sentence, but. If this is for children, caregivers should appreciate that reminder. I'll tuck it in after some fun examples so it doesn't scare anyone.
[00:19:45] Tom Adams: This is beautiful. It's just so good. Not not just this answer, but how you're, how you're processing the fact that, uh, an LLM just creates thinking based on a single prompt. I mean, it's fascinating.
[00:19:59] Ryan Neimann: It is fascinating. And number 20 was. The balloon is a potential choking hazard, this warning, especially for small children. So, so we, that was good. went through the 27, to choose the best com. Then it gives me suggestions to choose the best completion. these following elements like floating or deflated or popped or a, or a decoration. But I thought of all of these, the deep seek was the, the most fascinating of all of them.
[00:20:31] Tom Adams: Yeah. that's really good. I, I love the fact that, that it shows really simply, um, how LLMs only model is, what's the next most obvious thing I should do? And it processes that depending on how, how much has been built into that particular LLM model. I mean, there's so, it's so fascinating to see that unfold like that.
[00:20:53] Ryan Neimann: I agree. I agree. Well, much like Mike is diving in you, you start to transition this to how does it have some maybe business application or
[00:21:02] Tom Adams: Yep.
[00:21:03] Ryan Neimann: your own personal productivity you know. Tom, you and I have the ability to really think of this at enterprise scale and level and how these different LLMs respond affect how effective they are for users.
And so this is more focused on, how do I improve personal productivity, really those elements of increasing my AI literacy. I'm going back to this brand new chat GPT session, uh, that I've, that I've created and. Mike, I think you had asked last time, you know, how do I get a a, a trait, or how do I get custom instructions into my chat GBT to improve some of their prompt outcomes?
[00:21:46] Mike Richardson: Yeah.
[00:21:46] Ryan Neimann: I took the liberty of, of, uh, going to my settings and just kind of reviewing them. One of the things I always point out is to make sure that in your data controls, the improve the model for everyone is off. if you wanna make sure what you're sharing and putting it as prompt is not being used by the LLM, if the LLM offers it, which I know chat GBT does, make sure you turn that off.
So that's, that's number one. It's just kind of a, a good reminder for everybody that's listening.
[00:22:15] Mike Richardson: a reminder, everybody go go to the YouTube. Version because, uh, Ryan is navigating from screen to screen to screen right now and showing you exactly what buttons he's talking about and how to turn 'em on, how to turn them off,
[00:22:27] Ryan Neimann: Thanks, Mike. So I'm gonna go to personalization and now I'm going to custom instructions. And this is quite interesting. So remember we talked about you control certain things? Well, you control this, so you can give yourself a nickname. You can tell the LLM what you do. here it talks about, uh, a small batch home, sour dough baker, for instance, as an example. I'm gonna focus in on this element right here, which is what traits should GPT have? And we talked a little bit earlier about what I had nicknamed the Magic Prompt, again, a little bit. I'm gonna show it here in a second. Uh, Tom, I think you're gonna post it for our listeners
[00:23:07] Tom Adams: Yes.
[00:23:08] Ryan Neimann: to, to the magic prompt, and through cultivating a number of different suggestions, kind of in that 2324 era, I came up with this prompt. Recently when I found out that OpenAI had a prompt improver, I ran it through that to even give it more refinement. And it's important to note that you can only put in 1500 characters. So this prompt that I'm gonna show you is really quite dense and rich, and the words are all purposeful. So here it is. I call it the prompt completion guidelines, or PCG for short. I won't step through every element here, but there are a few things I wanna highlight. One is that it will stop and ask you three questions and create a three to seven item checklist, and will start with this order. You'll see there's an order number one an order number two. In that order. Number two, it says, yep, I'm gonna take the first response, take into consideration some elements, and then give you a response too. And then provide a final summary. And in that final summary down here, it will how it refined the prompt, shape the response, and adhere to these prompt completion guidelines and the considerations that informed response to. So think about that. This, this prompt will. First start off with a checklist, ask you three questions, come back with a first response, and then on its own can take into a number of considerations and then come back with a second response. But then also tell you why and how it made improvements to its own response. That information can be really helpful, especially when it's something important that you're working on. So what I did is just cut this, and now I'm gonna paste it right into my trades. And I'm gonna hit save.
[00:25:05] Mike Richardson: So there it is. So
[00:25:06] Ryan Neimann: I'm gonna hit
[00:25:07] Mike Richardson: watch this on YouTube. It's happening before your very eyes.
[00:25:10] Ryan Neimann: before your rear eyes. Exactly. So these custom instructions, so what I'm gonna do first is I'm gonna turn it off for enable for new chats. All right? So you can do this too. So once you put this prompt in, if you're like, well, I don't need, it's for this particular exercise, you can turn it off. I'm gonna do that. And I'm gonna go in and let's just start a, a prompt. How about I'll just do research. There we go. it gives me some. suggestions. And I'm just gonna take, so I'm gonna, yep. Let's do research trends in artificial intelligence that seems timely and appropriate. So I'm gonna click go and, uh, it's saying, thinking it's summarizing AI trends, recent AI articles, and hopefully this happens pretty quickly for our, for the benefit of our listeners.
Do we have any hold music here? Anything?
[00:26:04] Tom Adams: uh, I could start singing of course. While it's,
[00:26:07] Mike Richardson: We're doing this unrehearsed
[00:26:08] Tom Adams: yes, there's no,
[00:26:10] Ryan Neimann: rehearsed
[00:26:11] Tom Adams: yeah.
[00:26:11] Ryan Neimann: you'll notice actually, if you think about when you first maybe tried an LLM if you haven't used chat GT for a while, what, what came back was fairly quick, and this is really taking quite a bit of time to think through an appropriate answer more than I was hoping for right at this moment.
[00:26:28] Tom Adams: Well, we did say in our show notes that this is raw and unfiltered, so we're we're getting the raw, the raw goods.
[00:26:35] Mike Richardson: this is, this is the free version of chat, GPT, but is this chat GPT five.
[00:26:41] Ryan Neimann: that's a good question. So,
[00:26:43] Tom Adams: I.
[00:26:43] Ryan Neimann: five, uh, I, I, it doesn't explicitly say
[00:26:47] Tom Adams: Now this is probably four one.
[00:26:48] Mike Richardson: Okay. Okay.
[00:26:49] Ryan Neimann: is it? Okay.
[00:26:50] Mike Richardson: Right.
[00:26:50] Tom Adams: Yep.
[00:26:50] Mike Richardson: Okay. Anyway, yeah, keep going.
[00:26:53] Ryan Neimann: All right, so it says, here's a crisp source-based snapshot of the biggest AI trends right now, as of September 5th, it goes into some, some detail on those. All right, so let's just see. Let's just see what happens when I turn on magic prompt. And if I go to personalization, custom instructions and enable for new chat chats, and I hit save, I'm gonna create a new chat. I'm gonna say research. I'm gonna say research ideas for my science project, and let's see what comes back. Oh, quick comes back and you'll see here how how the prompt response acts differently
[00:27:38] Tom Adams: Completely different. Yeah.
[00:27:39] Ryan Neimann: So it already came back with a six point checklist. It refined my prompt and it's asking the three clarifying questions.
Now, we don't need to actually demonstrate, I do some easy things like just do a one dash and start to answer the question. You have to hit, shift enter to get a, a second line without hitting enter. answer the second question and I'll do three dash and a answer The third question, and what comes back is a much more refined and tailored response for what I'm trying to accomplish.
And in this case, think about my science project.
[00:28:15] Mike Richardson: So it's using, if you just scroll back up, Ryan, it's using the structure of your magic prompt that you put in settings to create that checklist there. Quick checklist. Those six steps, it's going to take you through. Is that correct?
[00:28:34] Ryan Neimann: Correct. Yeah. It's, it's called a custom instruction. And, and they're, they're different terminologies and different language models. In Chacha, BT they've kind of changed the name to what traits do you want the LLM to display? Um, but same effect in
[00:28:50] Tom Adams: Yep.
[00:28:51] Ryan Neimann: create the prompt and then it says, okay, what are my custom instructions?
And now I'll go to work. but completely. Changing the way that you work with an LLM and in many cases, ensuring the type of response you get back is, is impactful.
[00:29:05] Mike Richardson: Because we talked about this in the last episode, didn't we? Everybody, we, we reminded you of what we've always said about technology and databases and systems. Rubbish in, rubbish out. And so you might be, you might be tempted here, and I think, I think Mark said last time that he tends to be a bit, sort of trigger happy. He just wants to hit return, right? Get something out rather than spend more time putting in, in the first place. So what you are evidencing here is the more I put in, the more I get out. So it's worth, it's worth spending an extra minute or two or three or five putting in, and suspending your, your desire to get out just for a few minutes.
Is that right Ryan?
[00:29:48] Ryan Neimann: think it is, it's, it's part of that AI literacy and how to effectively work with, uh, a large language model. We've demonstrated that there's any number of models out there that accomplish. Different tasks specifically, or respond to a, the very same prompt in very different ways. But then there's also ways to elevate the way you're working with that lmm, LLM, when things start to matter, like your, your personal productivity or you're working with, with, uh, you know, a business task that you have at hand. all leads to. Which LLM is the right one for particular large scale solutions, right? Those are all decisions that need to be made and, and those, uh, types of architectures. But this is really just focused in on personal productivity and AI literacy. I.
[00:30:37] Mike Richardson: Beautiful.
[00:30:38] Tom Adams: Yeah, it's fabulous. Fabulous. So good. So good. so the, the, the thing that I, I think is also helpful here is if you put this into your. Your baseline setting structure, it's gonna force this every time. Uh, and so there are times that you, you may go, uh, I don't wanna deal with you. I just wanna be Mark Red Redgrave every so often.
Just to go trigger happy. And there's nothing wrong with that. Um, but I think it's, it's helpful to know that this, this. Using this kind of capability in your, um, in your settings, forces you to be more efficient and effective at prompting it just forces it, right? That's what I like about it.
[00:31:20] Ryan Neimann: fact, Tom, you just reminded me. So, you know, I just to keep everybody uh, uh, on the same page, you can
[00:31:27] Tom Adams: Yep.
[00:31:29] Ryan Neimann: so you know, if you don't want, that prompter, you can turn it off. But I found the Mark Red grave setting. And have it act as a cynic or a, a critical and sarc response every time.
[00:31:47] Tom Adams: Yep. Every time, which, which will be perfect. Um.
[00:31:52] Mike Richardson: have both of those activated, right? You can, you can activate the, the personality, which again you're showing on the screen here, and you could leave your. Your prompt, uh, structure, active, right, and you get a combination of both now, is that correct?
[00:32:06] Ryan Neimann: That's right, We've solved Mark's Mark's problems. So where is he? Where?
[00:32:11] Mike Richardson: are default, cynic, which is critical and sarcastic, robot efficient and blunt listener, thoughtful and supportive, nerd, exploratory, and enthusiastic.
[00:32:24] Ryan Neimann: But, but keep in mind, you know, Mike, I think you brought this up on, on our last episode, we were talking about, well, where's kind of that default setting and it's cheerful and adaptive,
[00:32:34] Tom Adams: Yes.
[00:32:34] Ryan Neimann: Mark didn't want.
[00:32:36] Tom Adams: Right.
[00:32:37] Ryan Neimann: Right, exactly.
[00:32:39] Tom Adams: Yeah. That's beautiful. That's beautiful. I think it's just such an important way to see and. Uh, and be aware of this additional layer that we can all use to amplify our effectiveness with using the tool, uh, with using the, the, uh, capabilities that OpenAI has. Um, there are also, if you use other LLMs, and I happen to subscribe to all of them, uh, I've been testing this one in like Claude and that, and it, it's pretty close to effective in all of those other platforms as well.
So Gemini has capabilities. Perplexity, which kind of does different things. but, it's really forcing you to think differently about how you're prompting. So, Mike,
[00:33:23] Mike Richardson: you've, you've, you've been copying and pasting that same structure that Brian just showed us, and you've been
[00:33:28] Tom Adams: Yep,
[00:33:28] Mike Richardson: into the settings and those different LLMs, putting it
[00:33:31] Tom Adams: yep. Yes.
[00:33:32] Mike Richardson: it, leaving it on, and
[00:33:34] Tom Adams: Yep.
[00:33:34] Mike Richardson: it.
[00:33:34] Tom Adams: And trying it now chat. GBT really is the leader in the regard to, um, really making this user friendly. I think they do a really good job. So those extra settings that Ryan just talked about, which is the cynic and the, uh, it's really good at that kind of stuff. It really has figured out, uh, OpenAI has really figured out, um, user interface stuff that makes it so much more compelling.
Some of the other ones don't do as good a job at that. But there is the capability to add these super prompts into it. Um, another thing, uh, uh, Ryan, that I might add to this is there is in all of the tools from chat BT to Claude, to perplexity, um, gr all have these, they have a structure called projects.
And in projects you can actually. Use this prompt within a project. So it's a way to, to hack not having to use it all the time. And so if you put this into a specific project, you, you have the ability to actually add in that capability, um, that what you call your super prompt, into a project. And thus everything in that project responds to that requirement then.
[00:34:46] Ryan Neimann: Yeah.
[00:34:46] Mike Richardson: you said they all have that capability and, and sometimes that, that's called something different.
[00:34:51] Tom Adams: Yes.
[00:34:51] Mike Richardson: so. For instance, in copilot, think it's called Notebook or something like that, where you can go
[00:34:57] Tom Adams: Yep.
[00:34:58] Mike Richardson: and, now that you're operating on a, on a inside of a project
[00:35:02] Tom Adams: Yes.
[00:35:03] Mike Richardson: in this case notebook, it's accumulating and compounding.
[00:35:07] Tom Adams: Yes.
[00:35:09] Mike Richardson: uh, on that, within that container of that project. I
[00:35:13] Tom Adams: Right.
[00:35:13] Mike Richardson: forever if you, if you keep asking it to refresh and deeper and broader and keep going.
[00:35:19] Tom Adams: Yes.
[00:35:19] Ryan Neimann: add one more for you. If you don't know where to find the settings, you can literally cut and paste that prompt completion guideline in your prompt, so
[00:35:30] Tom Adams: Yes.
[00:35:30] Ryan Neimann: I wanna research a science project, the prompt completion guidelines, and it'll follow it. Then with that prompt.
[00:35:38] Tom Adams: Yeah.
[00:35:39] Mike Richardson: And we're gonna put a link in the show notes, Tom, where they
[00:35:41] Tom Adams: Yeah, show show notes will go to our new prompt and circumstance GitHub account, where you can pull the raw data file right from there really easily. So you just copy and paste it from there. So.
[00:35:54] Mike Richardson: everybody, um, Tom shared that with us just before the show sent an email around. So I, I wrote back saying, Tom, please, in plain English remind me what GitHub is again. tell us what you came back with Tom.
[00:36:09] Tom Adams: okay. Well, I, just gonna show you 'cause since we're showing things today, I'm gonna show, 'cause this is actually showing Ryan's. Tool at work. So you'll notice I asked it to say for a newbie, how would you define GitHub using a metaphor or some other thing? Like GitHub is to software.
What? Blank is to blank and then it,
[00:36:30] Mike Richardson: that Tom was being quite polite, he called me a newbie. Thank you, Tom.
[00:36:34] Tom Adams: yes.
[00:36:35] Mike Richardson: I'm grateful that you, that's all,
[00:36:37] Tom Adams: I,
[00:36:37] Mike Richardson: me.
[00:36:38] Tom Adams: I did the best I could, but notice it comes back with checklist. Define GitHub. Using metaphor, keep it simple. Newbie friendly. Use clear X as to Y. What A is to B. Provide three to five variations. Avoid jargon, keep it relatable. It refined the prompt. It asked me three clarifying questions. Um, Ryan, you'll notice, uh, I didn't give it numbers, I just answered and it's quite fine with that.
Um, and then it gave me the answers. GitHub is to software. What Google Drive is to documents. GitHub is to software, what a fridge is to leftovers. Uh, and it gave a bunch of really cool responses that I then sent back to Mike and Mike. Did that help understand what GitHub was?
[00:37:16] Mike Richardson: I, I wrote back to Tom saying, now you are speaking my language. Thank you so much indeed.
[00:37:23] Tom Adams: Right. And so I think it's, it's what, you saw in a really simple way is that it, it really gave me a chance, even on the simplest of requests, it forced me back into clarifying the prompt and being really clear. And now I got great answers back instead of. Potentially a random set like Ryan showed us at the beginning where one simple question gives you a multitude of different answers.
So
[00:37:49] Mike Richardson: you've just caused something to pop in my mind, uh, guys, and that is this concept of clarifying questions that you want it to ask you. For you to clarify the prompt. Of course, that relates to what we do in peer groups. If any of the listeners are members of my peer groups or other peer groups, you know, it's before we rush to, you know, judgment about a particular issue, challenge, problem, opportunity that somebody's got. And we get very solution oriented.
[00:38:18] Tom Adams: Yeah.
[00:38:19] Mike Richardson: to understand the problem first. And so we go through a phase called asking, clarifying questions. And sometimes we might spend most of our time asking clarifying questions to really zero in on are we working on the right issue, problem, challenge, opportunity? Do we fully understand it?
What was it? Einstein said something like, if I've got 60 minutes, you know, to, to work On something, I'll spend the 55 working on the problem and five minutes working on the
[00:38:48] Tom Adams: Yeah. Yeah, it's really good. Well, that was fabulous, Ryan. That was really helpful. And again, we believe that probably watching the video is gonna be more helpful than, uh, just listening to it. But I believe even listening, you'll be able to figure some of that out as long as you use the show notes to get access to that super prompt.
So, , let's wrap then with, uh, what can somebody do with this? And Ryan, maybe start with you. What's a, what's a next step anyone listening can do based on what we've talked about today?
[00:39:16] Ryan Neimann: Yeah, I always recommend go with what you know, right? Start with something that you're working on. Use the, uh, super prompt, the magic prompt, uh, you know, try it out. Uh, but it's always. I think best to start with some context or topics that you know, and that gives you the ability to see, you know, is this right?
Uh, is it hallucinating or not? The, the prompt completion guidelines are intended to try to eliminate as much of that error and emission or hallucination and. Cited examples of what you're looking for, but that's a great place to start. And uh, to Mike's point, have it challenge it. Say, well, what is a contrary view to this, if you will, and start to explore your ideas.
[00:40:03] Tom Adams: Yeah. That's fabulous. Mike, do you have any, uh, other recommendations for what to do this next week listening to this?
[00:40:09] Mike Richardson: I'm gonna do this with my forums, uh, in the next cycle here over the next few weeks, a month. And, um, I'm gonna take some inspiration from the book, the AI Driven Leader, of course, uh, he recommends that Jeff Woods, the author, recommends, look, choose some low hanging fruit, choose a business process. Choose a task. Choose an element of your workload where you could sort of triangulate a better, faster, cheaper approach using AI in some way, shape or form. could find a much more efficient way to get something done, a much more effective way to get something done much faster, uh, better, cheaper way to get something done. And, you know, his recommendation is don't overthink it. Just create a long list of ideas. Come up with the short list, choose one. Lean into that one thing and start to, to ta really tackle as a project. Now, how could I tackle that one thing using ai? And you're probably not gonna come up with the right answer, right out of the gate. You're gonna try this and try that and try the other, and maybe 15 days later, 30 days later, 45 days later. Especially if you mobilize a bit of a task team and get your IT person and your marketing person or whatever it is around the table, you can create a breakthrough with that thing. And all of a sudden work load burden, has been liberated and you're now getting back, you know, some bandwidth to apply to more strategic things.
[00:41:51] Tom Adams: Beautiful, beautiful. Love it. mine layers into both of yours, but maybe. Uh, really pragmatically, which is something that Ryan was doing, was showing us how, how he gets around. If you're gonna be using chat, GBT, copilot, Claude, whatever, you're gonna use Gemini, get to know the settings. Like take, take 10 minutes and dig inside and go, is there personalization?
Where is it? How do I find this? Because so often we use like 10% of the capability because we don't know what Ryan just showed us, which is if you just do this over here or I want a cynical response, know where that is so you can actually adjust your toolkit to give you a better result for you, not just so that you get the magic of.
[00:42:34] Mike Richardson: Yeah.
[00:42:34] Tom Adams: Of the RESO response, like actually figure out how these tools work. I have found that extraordinarily helpful to me. Once I play with the tool, one of the first things I do is I never read the instructions, but I always go fiddle around and find things under the hood and I go, oh, if that's there, what else is there?
So it's really cool.
[00:42:53] Mike Richardson: me, it reminds me of the early days of any new technology, any software, right? You know, what you find is, oh, this piece of software has these buttons, whereas this one doesn't. There's a different set of buttons and you wish, oh, I wish, I wish that had this and that, and vice versa. So we said last time, everybody. Keep an inventory of prompts
[00:43:11] Tom Adams: Yep.
[00:43:12] Mike Richardson: because if, this particular platform that you want to use, because, because, because for this particular application doesn't quite have those settings, there's nothing to stop you being able to go over here to a Word document or a OneNote or something. And as Ryan said earlier. Copy and paste the super prompt,
[00:43:30] Tom Adams: Yep.
[00:43:30] Mike Richardson: and paste. I want you to act like a cynical, you know, mark Redgrave, or whatever it is, right?
[00:43:37] Tom Adams: Yeah.
[00:43:38] Mike Richardson: you can always have that inventory for yourself and achieve the same effect as you would in some of the settings.
[00:43:43] Tom Adams: Yeah. Beautiful. Well, gentlemen, thank you, uh, Ryan, uh, you, you carried the load today. So thank you to you for, uh, showing us around and doing some of that advanced stuff. And, uh, hopefully this was helpful. There'll be full show notes. Uh, check out our show notes. And gentlemen, we will catch up in a couple of weeks.
Thank you again.
[00:44:02] Ryan Neimann: Thanks everybody.
[00:44:03] Tom Adams: Cheers. Bye.
