AI for Work

Welcome to Building Better Developers, the Developer Nord podcast, where we work on getting better step by step, professionally and personally. Let's get started. Well, hello and welcome back. We are continuing season 20 where we're just mixing it up here. We've got our special topics and our interviews. This episode, it's going to be a new interview. We're going to be speaking with Ali Abbasi and we're going to talk about AI. I know that's maybe something you haven't heard about, but it's got a little bit of a buzz and he's put together a site that I think will get you, if you haven't been using AI, to think about a little bit differently. And if you have been using it to think about it a little differently, find some ways that maybe you can make more use of it regardless what your focus is. I know there's, depending on what your real focus, your daily career is, there's going to be certain things that you feel or understand are the strengths of AI. But you need to realize there's actually quite a few strengths to it, a lot of ways we can use it, and that's what we're going to talk about next. So let's jump right into our conversation. In the world of AI, everybody's heard about it. It's a cool thing. It's the buzzword everywhere. But today, instead of AI, we're going to go talk to AA. We're going to talk to Ali Abassi about AI and his site, and we'll talk about it. There'll be links and show notes, but AIforwork.co. It's one of those things that's like, we've all played around with it. We've all heard about it. And so we're going to get into it in a conversation about how do you really use this stuff and where is it at? It's sort of new. Where is it going? And so we're going to talk to what we're going to call an expert on it. And with that introduction, I'll let you introduce the rest of yourself. So what's your background? Where you come from? And welcome to the show, Ali. Awesome, Rob. Thank you so much for that intro. So my name is Ali Abassi, and my background has always been in sales and marketing in the D2C space. When ChatGPT came out, I became obsessed about using it in my own workplace, but I track my time meticulously. And I found that after using ChatGPT, I was actually spending more time trying to get work done because I kept having to edit these very vanilla answers and ultimately ended up taking more time and the productivity was not gained. But I knew the promise was there. So I figured out a prompt framework that allowed me to just get some of the best content I can out of ChatGPT. And then just over a period of time, I built thousands of prompts that use that same framework for every field. And I built a tool called AI for Work that helps anybody at any level get the most out of ChatGPT with prompts that are designed specifically for their job. That's a nutshell. Okay. Well, let's start with that because I think anybody that's playing around with AI has a little bit of an idea of prompts, but give us a little bit, I guess, just even for that, give us a little bit of a definition. What's a prompt? And then as you talk about how do you, what would be the difference between, I guess, a simple prompt versus refining that prompt is what you've done with the site. Yeah, totally. Okay. So a basic prompt is a request that you would have for ChatGPT to have some sort of output. And then based on how you want to refine that request or how specific you want that output to be, you would refine your prompt. The challenge that most people have when they're using ChatGPT is they struggle to find a use case that it's something that they could do repeatedly. And with code, that becomes a very good use case. Like you could use ChatGPT every single day for almost every task. But when it comes to marketing, you really can't. And yeah, and I think with all these large organizations investing in workplace AI, we're going to see a huge number of people who are in that space of I'm trying to figure out what to do at work with AI. Now that my company is investing in an enterprise open AI package or like a new Microsoft package, I need to figure out how to use it for my workplace. So what I did is I put together a bunch of different use cases. And then the framework that I built specifically involves the individual just requesting a document. So say you're a marketer and you want to have a marketing plan, you request a document using this prompt that you want a marketing plan. It asks you questions. Once it provides you the actual results, the difference here is that it evaluates its work. So at the end of providing you its response, it evaluates its own work, gives itself advice on how to improve it, and then asks you if you want to improve it. And over a period of two or three revisions, you end up with very high quality professional work you don't intend it. So how does it, I guess this maybe begs the question, if it evaluates its work and there's more that it can do, one, why didn't it do it the first place? And then two, is there some sort of a feedback kind of loop that it's working with with the requester basically, whoever's requesting the document? Yeah, great question. It gives you the option of providing your own feedback. But the real question is why didn't it do it the first time? And the answer is I don't know. I wish it did. But yeah, based on what I've tested is using this refinement and self-evaluation step, it allows you to add very custom criteria to it. So like one of the criteria is point of view from an industry expert. So while it's doing its evaluation on say your marketing plan, it takes on the role of another expert in marketing to critique the work, which adds a little bit of a unique perspective to it. So let's say, I think this is where people struggle. And I think a lot of people that play around with chat GPT in the AI world are like, hey, I've got I need some content, for example. OK, I get it. I write a paper. Boom, it wrote a paper for me. And then it's like, OK, like you said, you end up looking at it, go, OK, well, here's like a start, but now I want to rewrite stuff and I got to change stuff. So how do you what would be like an example of how you would take that and refine it? Like, let's say I just start with something very simple. Like I want an ad for my widget and it gives me something. You know, and I tell a couple things about my, you know, my bright, shiny widget or something gives me something back. What is like, what is sort of walk a little bit about through that process of how do you refine that? How do you now say, OK, yeah, that's cool, but not cool enough. I need you to do it with more gusto or whatever it is that. Yeah, for sure. That better iteration. Yeah, I wouldn't say there's a perfect process, but in my experience is when you when you give it enough input through the question period, one of the areas that I would put for anything related to marketing is just like a target persona that I'm trying to go after and the attributes there. Like what are their dream states? What are their pain points, their psychological pain points or like why they would want to? What are they actually buying from the as a product? And then use that information to have it as part of input and then to refine it as part of the prompt, it gives you one option where you could just give a general feedback. And then in that, I would just basically request how I used it. I would ask like, how did you use these references I provided in your in your final work? And it generally fixes itself or it gets it to the point that's close enough. So this doesn't end up being perfect, perfect, where I would say, hey, use this prompt and then submit the submit the work. But I would say it gets you much closer. Yeah, much, much closer. So is your does it is there a way then for you to like, as you said, sort of weights the prompts as a way that you can go in and adjust that and say, OK, like I talked about, you know, my my avatar or my persona, but I really wanted you to spend more time focused on making it professionally sounding or something like that. Is there a way that you could sort of tweak to come back? Yeah, so the way that the prompts work is they basically become a co-pilot for the individual employee and it always gives them a total of I believe it's four options. Like the first option is, oh, do you want me to improve my prior work with the new advice? Two would be like, do you want me to ask more questions to personalize it better? Three is, do you want me to be very stringent in my credit in my critique? So really to give it even more criticalized, so you have like that assurance that it's a good document. And then four is, do you have any general feedback? And then using these four options, it simplifies the process for an employee using it because it like, yeah, it limits the number of options, but also gets them to the end result, which is a higher quality document that that utilizes all of the things that they that they input. So now with the prompts, and I think you said there's thousands of prompts that you've created, you put together, what gives the prompt value? What is it that I, and I've noticed in this, anybody that, again, there'll be a link in the show note, go check out the site. You can see where it is a very nice way to sort of navigate your way into what is it that you really want. And then, and there's a little video, I recommend you check that out a couple of minutes, walk you through this, but outside of that, like in building this is what makes a prompt a good prompt, basically. Yeah, for sure. I think there's use cases for all sorts of prompts, including like single line prompts, a yes, no question type of prompt. But in the scenario of AI for Work, I've gotten so much feedback from users who are using it on the daily basis now. So for me, I would say a good prompt is something that an individual would use repetitively, that genuinely would save them a significant amount of time in their work. I think that would be a high value prompt. So like an example would be a priest who reached out and asked if I create a prompt for writing a sermon. So we did. And I see on my end over here, it's being used multiple times a day. And I tested it myself just to see what that process is like. And it is a fantastic job to get all the ideas onto all the ideas down. So I think usually when people think of a prompt, it's like, you know, a couple words or something very short. So can prompts be, it sounds like, can you put quite a bit into those prompts so that they're almost more like a, it's not a couple words, it's like a paragraph or something like that that's really, that's helping it along? Yeah. So I tested out so many different ways of doing this. What I found to work the best is if you write your prompts in JSON format and have parameters like prompt, rules, steps, you're able to do quite a bit, specifically on GPT-4. So yeah, I would say like if you're trying to make complex mega prompts, if you just use natural language, you'll find that it drifts quite a bit. But if you have them in JSON format, it's magical how well it is able to understand that format and to respond to it consistently with a similar response. But also from a formatting perspective with prompts like, Microsoft came out with a framework called Guidance a while back that I don't think a lot of people read into, but ultimately it gives you the strategies of how to get a prompt to give you the output in the right exact format that you want. But yeah, it all relates back to JSON format, which was quite surprising, but it works really well. Well, I guess that sort of makes sense because then you're helping it, instead of it having to translate your language or your English essentially that you use or your words you use, now you're back into something that's more tag based or it's very specific. Like, hey, this is a keyword or this is a rule, things of that nature. You said that you've got all these users and they're doing prompts. Are you assessing your own prompts as you go and looking at what's out there and taking that feedback and then massaging them and building those out? Yeah, that's exactly correct. So right now, my focus is on the area that I'm trying to learn as much about as possible is in the realm of AI agents. So AI agents and virtual employees, I really think this is going to be the growth area in the AI space. So I think the next stage after chat GPT is going to be interactive chat bots where it's verbal, you get feedback, like immediate feedback, just like chatting with chat GPT, with voice. And I think after that, we're going to go into AI agents where they complete tasks for you and connect with all sorts of applications. So like, say you have an HR agent as an employee and you want to do onboarding, that HR agent, which would be virtual, would be able to connect to Slack, set up an account, set up a Google workspace, send you your documents, all that stuff. I think that's the area that I'm most interested in right now and focused on. And AI for Work allows me to get a better idea of how individuals are using AI in the workplace. These users are all people who I would say are still early to it. Yeah. Yeah, and I think that is something that it is a great way that if you haven't tried it out yet, I think that's exactly the kind of site you want because it sort of gets you going in the right direction instead of just, here's some text, start writing something and see what happens. It gives you something that I think is particularly because of the way you categorize them. So it's things like, hey, are you in sales? Are you in marketing? Are you a technical person? Are you a lawyer? Are you trying to write a sermon? That kind of stuff, I think helps, sort of it prompts the user even to say, well, here's some ways you can use it and then go through that process and generate it or generate your content. Now, are you seeing, I'm not sure how much you're seeing what's being used and how much is this is just in the chat GPT world, but do you see a lot, like my thought is particularly like from content, if somebody comes in and they ask for content of a certain type and they're sort of going through the same kind of prompts and then somebody else comes through and roughly goes to the same prompts, particularly in your case, you've got these very detailed prompts. Are you seeing that they're getting the same results or is it different results but same quality that you're running into? Yeah, it's going to be very different results but same quality. And the different results will just be because like the step one out of the prompt is the questioning step. So it asks five questions about the individual situation and depending on how you answer that, that's what the output would end up being related to. So yeah, that's where the differentiator would be. It would just be the input that the user has, but they end up with the same quality over and over again. Okay, because that's again, that's one of those, it's one of those challenging things because if you're writing code, you'd love to have it like get the same code every time. Or if you're answering math problems or something that there's a, theoretically there's an ideal solution for it. So you would hope you get the solution every time. Content or writing and some of those things where you don't want it to be the same. You want it to be a different one so that it's not, you don't feel like you've just written the same thing as the guy or gal down the street. Yeah, content is such a hard one to crack because everybody has such a unique style and what they like. So it's like, I think it's going to be hard for chat jibs to ever get it to be perfect. But I think the things that I use from a content perspective is just characters. Like I really love Harvey Spector from Suits and I just love the style of his communication. That when you write an editorial in his style, I wouldn't say it's perfect but it gives you a lot of ideas of what you could incorporate into your actual content. Yeah and then I would say some of the best prompts that I've used for content would be to ask the chat jib to provide the output in a table format and in each column would have in the tone of a different individual. So you could pick five or six people and then have all of them compared next to each other. But that's how I run, how I use it for content. I just use the tone of characters that I really like. That's awesome. I never even thought about that as being able to do it in a different voice or a different style. Because that actually, but it does lead to, you said initially you were spending a lot of time tweaking the results and that's sort of what got you into this. Have you found that you have made great strides in the amount of time it takes you to go from start to finish when you're generating content? Yeah totally. So now I'm able to, like at this point I hardly use the chat jib app anymore. Most of my time is spent within, it's actually in GPT for Sheets. It's an application for extension for Google Sheets and you're able to use chat jib within it and it allows you to work at scale. So I'm able to make like 2000 blog posts in the span of like 10 minutes by just doing a big drag and drop and having the formulas set up in them. It's something that I think like can save almost anybody an hour a day by just using GPT for Sheets. It's an awesome tool. So now as somebody that, as I initially talked about, is sort of a trailblazer of sorts and that's spent some time in it obviously, where do you see the, where do you see like some of the strengths and the weaknesses of chat GPT? Or actually AI in general even? Yeah I don't think it's gotten content done right? Like it's not, it's just not at a stage where if somebody could say hey let's just make an AI movie or let's use social, like an AI social media generator to make content, the content just wouldn't perform. It just wouldn't end up being the quality that is engaging. So it hasn't cracked human creativity to that degree yet. So obviously that's the big con. And then the other thing would be that it's just really challenging for a lot of people to figure out what to do with it. They don't understand the use cases or fully understand how it could implement, like implement, how it could be used in their specific work or even in their life. So I think that learning curve is still there. It's going to be experienced. But I just don't think it's there yet. So do you find yourself being sort of a, you know, an evangelist for AI from where, because obviously you found some use for it. Is that part of what you're doing is sort of getting the word out and feel like you're sort of telling people hey there is value to this and here's how you can use it? Yeah absolutely. I think from a business owner perspective, like if you were to just see a competitor use a basic prompt and get 25, 30, 40 percent more value out of it than an individual not using it, you would take a look at your workforce, right? Like at this point it's going to become a productivity conversation at first, but then it's going to become an economical conversation. Because like if you have a small business with let's say two virtual assistants, you could turn those virtual assistants into experts with a tool like ChatGPT. So I think it changes so many different things. But yeah, I would say like it's not hard to convince somebody to automate their work with AI. Like the chat portion of ChatGPT isn't revolutionary. It's a decision-making portion of it. It's the ability to say give me a yes-no answer based on this details and then you can automate stuff on that yes-no answer. Yeah I feel like that's the real value of it. Now have you seen it, it's like a lot of these technologies is they sort of like they hit the ground and then it's just like iteration, iteration, iteration. People are working on it, they're trying to improve it and of course you also have that feedback because you have users that are out there that are messing around with it and giving feedback into it. So in the short lifespan that it's been there and the time that you've been working, have you seen, what do you think about the progress that you've seen on it? Is it something that you see it's going to be like six months from now it's going to be crazy ahead of where we are now or you see it's like it's working at a pace where we're going to be able to just, we're going to be able to leverage it as it goes. How do you see it you know moving forward? Yeah I think it's like an early explosion moment. The models are going to get bigger, they're going to get smarter and they're going to, it's going to happen faster and faster and faster over the next six to 12 months. I think like, I think from a workplace perspective anybody who's working at an office who currently doesn't use AI at work for whatever reason they're going to in a 12-month period of time, the applications they use are just going to start to have them. So I think it's just going to become so common and I think the size of these models are just going to explode. There's going to be so many of them and so many different options like, so yeah I think the future of from a development standpoint it's been happening way faster than I would say anybody that I know has expected it to and I think the it's just going to keep getting faster and we're just going to start seeing new things come out rapidly. And we will pause there and we will come back next time around. We're going to wrap this up, we'll have part two of our conversation. As I think you've seen now, Ali's got quite an interesting, little different background. This is a perfect example of somebody who essentially had an itch to scratch and they created a solution for that. That's one of the things that we've talked about in the past and when you talk to some of the more like out there thinking outside of the box solutions, they very often come from this kind of a situation where we've got something that we just need to do. We've got a problem we need to solve so we solve it ourselves and particularly as a technologist our probably first way or first choice for a solution is an application which often means we can then take that at some point, turn it into a product or a side hustle. Something to think about and I'll let you do that while we wait for the next episode. But until then, go out there and have yourself a great day, a great week and we will talk to you next time. Thank you for listening to Building Better Developers, the Develop-a-Noor Podcast. You can subscribe on Apple Podcasts, Stitcher, Amazon, anywhere that you can find podcasts, we are there. And remember, just a little bit of effort every day ends up adding into great momentum and great success. Please allow me to take 30 seconds of your time to talk about one of the things we're really excited about for 2024. We are going to bring back our masterminds and you can check out technologymastermind2024.com or you can check out our mastermind at develop-a-noor.com. We're going to get our groups together, we've got applications open today, there is an early bird discount, jump in there, take a look at what we've got and make 2024 your best year yet.