AI Unleashed: ChatGPT Productivity and Work Efficiency Hacks

Over the past few years, a dedicated team has been immersed in building a Generative AI startup @ HAL9 The journey has been a masterclass in innovation, resilience, and learning. Some strategies paid off immensely (enter ChatGPT productivity hacks), while others served as hard-earned lessons about the dangers of over-investing in the wrong initiatives.

To share these insights, an upcoming webinar with Javier Luraschi will focus purely on learning—no sales pitches, just real-world experiences. It will explore the highs, lows, and everything in between of AI product development, offering practical takeaways for teams and creators navigating this space.

Key topics include:

1️⃣ AI Productivity Hacks – How tools like ChatGPT have redefined work efficiency.
2️⃣ Product Development Wins & Fails – The costly missteps to avoid and strategies that led to breakthroughs.
3️⃣ Creative AI Tools – Evaluating tools like MidJourney, Flux, and others: the game-changers vs. the underwhelming.

This session is designed for anyone looking to maximise their AI tools, refine product strategies, and learn from the lived experiences of others in the field.

Javier Luraschi is a software engineer with experience in technologies ranging from desktop, web, mobile and backend, to augmented reality and deep learning applications. He is currently the founder of Hal9 Inc, a Seattle-based company dedicated to making Artificial Intelligence truly accessible. He previously worked for RStudio, Microsoft Research and SAP and holds a double degree in Mathematics and Software Engineering. Javier is the author of various R packages like sparklyr, mlflow, r2d3, and torch.

Thank you to our Community Patrons

VisAgi

Change Angels

Businessmap Formerly Kanbanize

Women in Technology and Science Ireland

Find us here: www.agileleanireland.org

Transcript

Hey, welcome to today's session on practical hacks, how to efficiently and effectively use a AI. We would like to thank you, our patrons and sponsors and community partners and Business Man, formerly known as companies charging. Managers and women in technology and science. Ireland, who all show us great support and partner with us to deliver those meet ups and especially today's meet up and I would like to invite you formally now to our today's meeting which is charge PT productivity and work efficiency hacks and is presented. But Javier and Laura ski. Who is now based in Seattle and he is founder of Hal 9 Company and he specialises he, you know his degrees, mathematics and software engineering but at the moment he specialises in making AI artificial intelligence very accessible. All. And Javier will tell you that himself, but he's in the middle of a storm. Massive storm. So he mentioned to us that that his Internet is not the most stable. So hopefully, fingers crossed, everybody we won't have any problems. But if, if if we lose the heavier, please be. Mindful it's it's beyond our control and the last thing before we start the talk today is very packed. So we will ask you to to to take a note of any questions you. Ahead and hopefully at the end we'll have time for questions. If not, we will gather them and then we will follow up with Javier and we will have follow up session or just you know maybe newsletter answers to any questions you might have. So we apologise for it, but it's the session sparked. So. So let's let let's start. Have you heard the sinceres?
Thank you, Joanna. And can you confirm that you can see my screen? Alright, great. Yeah. So we have a lot, but you are already mentioned just a few seconds of overview of what's going on in Seattle. So Seattle it's in the West side of the United States and we just got like a tonne of wind. Since 2 days ago and you know like that's how the road of my you know how it looks like and that's the power outages on the left. So it's literally half a million people don't have power and the crew that was repairing power also got a tree fall on top of it. So it's, you know, I think it's gonna be all good but just. Giving you a. Heads up that there's a lot of power. Just in Seattle. So, fingers crossed, everything goes smooth and we have like, a super packed agenda for today. So I'll just spend a couple minutes on intro. Then I'm gonna try to cover all the content, which is a lot. I'll explain. It and then we have a few minutes for Q&A. I already got a few questions in advance, so I'll definitely cover those. But yeah, it's gonna be packed with information and the good news is that I'm hoping that you get a lot of value out of this. The bad news is that you might have to drink extra coffee while you're in this presentation, but it should be all good. All right, so a little bit about us. You have already interviewed me. I'm Javier. Hi, everyone. I'm my background is software engineering and mathematics, and I've worked at different technology companies and now we're working at a company called Hotline that we cofounded. My brother is my cofounder, and he's electric mechanical engineer. And we mostly have a lot of technical people in our team. So we're. A bunch of technical people trying to make chat app team easier and more accessible at companies like the ones that you're. App. And we're also part of the Paul Allen Institute for AI here in Seattle, which has like hundreds of AI researchers and the like. And this is just to give you a background of who we are. And I think what you should take from this slide is that we know a lot of AI talking about generative AI and CDT and the like, but for this. Presentation I want to start from the very basics, so we're not going to talk almost anything about what we actually do on our company. We're gonna talk about how do we use chat DPT and the like in our startup and in my day-to-day. Job and the first thing that I want to mention, and again I'm starting from the very beginning. But you know, if you haven't heard about ChatGPT, you should definitely start checking it out and the like. And the best evidence that I have is if you search Google Trends chat, if it is getting searched 4 times as much as Excel. So just think about that right? You know, when I was a kid, Excel was just starting and I saw businesses adopting it and changing the way that they do financial projections, the way that they do planning, the way that they do their taxes, they that they do strategic decisions and building and construction and scrum and. Right. So think about that and JPD is 4 times more used or search at least than Excel. So that gives you a sense of, well maybe this is a random technology. I shouldn't look at it. Well, that's really not the case like it's getting, it's getting a lot of attention and I think it has a lot of potential. And again, zooming down into that, you might think, well, I'm an accountant, right? Or I'm a scrum master, like, you know, it doesn't apply to, you know, charity should not apply to me, right? Or I'm in healthcare. That shouldn't apply to me. This is a survey that has been started with that got started on 2023. So we have now the years of adoption of chat DPT of 2023 and 2024. And what is interesting is that literally all the industries are using chat DPT to some degree finance and accounting and legal. Were the two industries that were lagging, so those started using chat DPT on 2024 and there was less use on 2023. But basically at this point it is very well used across multiple industries. And just a good reminder, right, I really don't think that there is an industry that should not be using it and we can talk hopefully some use cases and the like, but it is, there is no excuse like everyone is using chat DPT. Now they. Oh, sorry. Yeah. Well, I I didn't mean to click that link, but that research was from the Warton University. And it's a well known source. Now the the interesting thing is, you know, if you only put in 2 minutes of attention in in this webinar, I think this might be the most important slide. What I wanted to convey on this slide is to say. I really I really believe that if you know if you're not using you know charge PT, you know, like your job potentially could be at risk, right? And it doesn't matter where you work, it could be on engineering, it could be on sales, it could be on healthcare, it could be on legal, it could be on government. I really do believe that all our jobs are at risk or you can see it as also an opportunity and. Like if if we're not using chat DPT at work, it's probably a red flag. But the other on the other side, and it is equally as important. If all we're using at our jobs is using chat DPT and answering questions with chat DPT, that's also really bad, right? It's you know we cannot rely on chat DPT to do our job. It doesn't matter where you work, right? Like engineering research, government or healthcare. If your if you go to the doctor and you ask and you give your symptoms. To your doctor and the doctor. Copy paste that into chat app and gives you the result. You know that's probably a doctor that I wouldn't recommend and that's that's not the way of using chat DPT so. Really, the challenge that we have in front of us is we have a very important technology that we should all be using, but it's also, you know, it takes, it takes some understanding to figure out how to use it properly and where it's the needle really in between, not using it at all and using it for everything. Both. It is my point of view that both are wrong. And we need to find exactly what's in the middle and how the middle is moving. Over time, so say another way, that example what we're gonna look in this webinar is we're gonna start with the simple use cases, the types of cases or hacks that I use where I use chat DPT for literally everything, right. And you might think like chat DPT is great like I should use it for everything. So smart and you know, so competent. And you know, he tells me all my answers, all my question. That's that's. That's the first part of this, this webinar, but in the second part we're going to look at all the problems that chat DPT has, right, and we want to understand like, what are the things that it's not good at. And then you know the third part of this webinar, we want to come back and figure out like after we understand, what are the things that child PT is not good at. It's like how can we overcome those limitations and you know, try to use it for everything, right. Ideally at some point in the future.
Sure.
Yeah. And you can think also of these. This same statement you can think about it as a road map adoption in your organisation. I think the last three slides I'm saying the same thing on different ways, but you can also think of each organisation is going to take some time to adopt ChatGPT right now I think everyone is excited and is figuring out what to do and how to use it. Everyone is on the very initial organic phase, but also throughout this webinar we're gonna be looking at example. Of. Again, like how do we move from naive use and and get us a lot of productivity gains in the short term to like understanding the issues to overcoming the issues. So this webinar is kind of like trying to cover everything of simple use cases, you know, caveats, you know, complex use cases. So you know, we're going to start with small steps, so I apologise for some of you. That maybe you're already a power ChatGPT user and maybe it's not as interesting like the first. The next 5-8 minutes, but we're going to start with the very, very simple use cases and that should get us everyone. Up to speed. So yeah, again, I mentioned the three things that I'm going to split this webinar in. You know, we're starting with the first one, which is just talking about productivity hacks and the like. And then we're gonna look at the more advanced use cases. So the first use case is use chat DPT. As smart coworker, you know think of chat DPT as hey. I have a small coworker. He might not be. Perfect. But let's just ask questions to chat DPT. That helped us answer those type of questions and I have a few examples here. All of these are real examples that I use on my daily life, so hopefully they serve as inspiration of how you can use GPT. So the first one is OK, I have a smart coworker that knows all the questions. In this case, I had to change the e-mail address of a coworker in Slack, and you know, like I didn't know how to do that. It's a bunch of settings. So what do I? I could potentially go to Google or spend like 5-10 minutes figuring out how to do the e-mail changes for coworkers. You know, I don't have to do that. You know, I don't have to go to Google and spend those 5-10 minutes. I can ask charge EPT. You know it gives me the steps and I get it done. Here the productivity gain is maybe like from 5 minutes down to like one minute or two minutes. You know, whatever they're like easy gains on the day that that do help my bottom line. Another case I was looking for investors of our company and I was researching one particular organisation called Greater Seattle partner. This. Same thing. You know, I could. You know, I could. I could do the search. I could read all their, you know, you know, Web page, figure out the correct things and like or I can just ask chat DPTS like, hey, what does this company invest in? Gave me the thing. I realised that it was not for us. Next one maybe it saves me 15 minutes. Right? 20 minutes of researching a company. And again, it's just like easy, easy. Things another one on this use case like this one is a little bit more tricky, so I won't explain the details, but I do a lot of engineering. We were talking about a technology called Kubernetes and we had a simple question right. And hey, like I just need that point of view, we got the answer and it was done. Now you can apply this to your. Job right? And you know, say, say that you're an accountant, right? And you're figuring out like, hey, what are the new regulations that I should be looking at? Just simple things give a couple of pointers. You're not trusting chat DPT 100%, but they're just kind of like sending the ball. Trucker worker. You get it back. You know, you get the answers done. So anyway. So that's. That's the first kind of like bucket of use cases that you can use charge. PTS. Just think about it as a smart, smart coworker. That answers some questions. The next one I use chat DPT. When I have a really, really hard problem. So say that you're solving something really hard on energy production, right, or materials or construction. In my case, it's engineering, right, software, and at times you're going to be very stuck with a really hard problem. Don't expect charge PPT to answer the problem for you. That's not what I meant on this use case, but use it as a way of getting unblocked and I I encourage this a lot to junior software engineers where hey, before you come with a question to me which is going to take my time. You know, I have to explain you and the like. Whatever question you have, you know, if you're stuck, ask ChatGPT. And again this is applicable to different roles and the like, but I'll show you a couple of examples. You know, so I was again doing fundraising for my startup. And you know, like ABC was telling me, it's like, hey, you know, like, you know, you need to, I can't underwrite the risk, and I don't understand it. And it's like one of those things that I'm not an expert on fundraising. And I felt stuck. So I asked the question, and I got a few, a few suggestions from chat DPT. And again, I don't expect charge apt to tell me how to successfully fundraise on my startup, which is really hard. But hey, I'm stuck, you know, here are the things that I've tried. Give me give me a few pointers. Here's another one. Sorry, skip one. I was preparing this presentation. And. And you know, like I, you know, I I was trying to group the slides into 3 topics and I couldn't find the last one. And I'm like, OK, I don't know what to do and you know, you know, again, I'm not expecting the answer from chat DPT, but it's just just give me options. It's like this is what I've tried so far. What are the things that I haven't tried that that might help me out? And again it just. Kind of brainstorming maybe 4 out of five are gonna be incorrect or not useful, but you only need one to be on stock, right? And that potentially could save you hours or days of work, right? And another one, this one is super technical. We do AI projects, we help companies and the like. This is very nerdy, but we were solving a problem with a neural network, which is an AI model. Replace that with whatever you want, right. You're doing accounting for for Ireland, but you're based on the US and the companies. Also has represented business in Africa. I don't know. It's just like super tricky. What type of regulations should I look at? It's just very hard technical question. I don't expect the answer from chat DPT. I just need to brainstorm and get some answers. Right. And the last third use case, you know that covers all the simple hacks that I have on my day-to-day. I like this one a lot. I see it less advertised on social media and the like, but this is great for people that are a little bit of perfectionists. At times I can I can dwell too long into one particular task, and I just want to make it better. So what is the fix? Is already a very clean cut, even making itself cleaner, which at some point maybe it doesn't really help you too much to spend too much time in tasks. So here are the examples that I have for you. So usually the prompt that I use. It's literally just improved and. You know, in this case I was doing like a social media post, right? It's like, hey, this is what, how 9 is it's add, you know, like, you know, blog posts, whatever. And. So it. You could argue that it's already mostly complete. It's like, OK, that's a reasonable blog post, but I just want to make it better and better. Might mean checking grammar, which judge if it is great at, you know, targeting a message for your target audience. You know, maybe you work with a lot of engineers, but you're more on the business side and you want to, you want to send an e-mail. That it's a little bit more technical and can do that translation not just from some business, speak to more of a technical speak, right. So I think it is great for all these things where you already have some text. And you know, it's it's kind of like 9090% done, but you still feel like, well, I need to improve it a little bit more, right? Like this doesn't feel right. You know you can just post the text that chat DPT and say like hey, help me improve this right and it will improve different things and then you can have that conversation of well like can you make it shorter, can you? Make it longer. You know, what about this? Can you change? Why? Here's another example. It's silly, but you know it saves me time. You know, I was just writing an e-mail message to someone that I care, and I was just congratulation congratulating them, you know, to, you know, to their family is such a meaningful event to cherish many memories. It was just like the closing line on an e-mail. But I looked at it and I'm like, why doesn't it doesn't feel right? So like is my grammar incorrect? Like what's going on there? And again what I did is just hey, here's what I have. It's 90% complete. Can you help me improve it? Right. And and I got like a couple of examples and it's like, yes, this one feels more personal. I like that one. Improve it, send it. That way and another example here, e-mail that I'm improving the first e-mail, it's ready to be sent potentially, but I read it and it's like, yeah, that, you know, something falls off, you know, like, what is it? And again, I copy paste it, put in chat, DPT, get a few suggestions and again this is not. And I I've seen, you know, and this wraps up this section, but I think what is important to take from this. Action is. Sometimes we think about chat Gpas. It should be a tool that saves me 20% of my work, right? So we expect we, you know, we could potentially expect that chat DPT. You know those 20% of our tax preparation right or 20% of healthcare diagnosis, that's that's not really the way that I see the productivity gains. It's mostly what I showed is throughout the day, if I can say. Like 10 minutes here, 20 minutes here, 30 minutes here. That's the 20% productivity gain that I can accomplish with these types of hacks. Is not rocket science, it's simpler things. It just helps me focus and get more a little bit more work done and be more efficient. That's pretty much it. It's not. It's not like the robots are taking over our our job or anything like that. But if you become an A proficient user of chat DPT, I feel very confident that you know your productivity at work is going to increase and it's little things right. It's not. It's not like life changing so far. All right, so we're doing OK in time. I know that we're going fast, but we're to the point where, OK, like I understand what Javier means by productivity. Gains are simple things. You know, like questions that I'm sending. Like, great. I get it. I'm going to try that GPT, and I'm going to get those productivity gains. Now what is very important. Just to understand. What are the caveats of Chad, PT and this is this is getting in the middle of? Our road map where? You know you should be aware of the things that can go wrong when using ChatGPT, so I'm on the next few slides. What I'm going to do is criticise chat DPT in many ways and then we're going to come back to try to fix those problems at the third part of this part of the webinar. Yeah. So the way to think about this is for those of you that have watched Space Odyssey and our company name, Hal 9 is inspired on Hal 9000 basically. Things went wrong on space odyssey and you know, at least you know they, you know, our belief. Is that you know, AI will need improvement and we will need to understand what are the caveats of AI and we'll have to fix AI, right? So in this case, we believe that how nine that we can make? Yeah, I better and a lot of the work that we do is related with improving AI and the workflows and so anyway. So first first problem and again. Gonna go very negative then. I'm. I'm gonna fix with positive ways. How we can fix this. So the first thing that you should be aware of is that if you look at the terms of service of charge DPT. JPD has the right to use your private private data to improve the model, so potentially you know if you tell a secret to chat with PT like hey, you know, you know like here are the secret secret nuclear codes, right? Or whatever, you know, potentially if a third party country comes and asks like, hey, what are the secret codes? From the US, right, they have the right to improve the model and pass those secrets to someone else. So again, we'll come back and figure out how to fix it, but you should be. Aware of that, by default, anything that you do on chat DPT, you're giving the right to chat DPT to improve the model and make it better for you and for everyone else which could potentially be a privacy issue. The second one. Oh, sorry. I didn't mean to click, but those are the terms of service. The second one is, you know, when you when you ask chat DPT for say. You know, like a private question, right? Like if I if I ask charge if it is like, great. Have you taught me that I can use it as a coworker? I'm gonna ask like, hey, what are the emails of my customers? Right. It's just a simple question. Should be able to answer it. And the reality is that no, it can it doesn't have access to private data it's. You know it's gonna give you some text that, hey, I don't have access to your private data. Like, that's not something that I do. It's like, OK, well, well, that's a problem. Like, if you don't know anything about my private data, how useful can you be? Then you know I asked the I asked the question is, is it good to water my garden? Right? Like is it good to have a well water lawn? And the answer that I got is like, yes, having a green garden, you know, it's great and etc, etc. And my point of view is if you care a lot about the environment, maybe it is not as important to you to have like a green. Beautiful garden in your front yard. Maybe you care a lot about water concerns. But in this case, deputies giving us an answer is being very factual, and it's telling us, yes, you should. Do it and. The problem here is that Chad VPD is biassed. It has its own point of view of the world, and at times it can give you answers that maybe are. Not necessarily incorrect, but they're biassed to from what JPT thinks is good in the world or not. Next one? So I asked Chad DPT. Give me give me the three book titles that Xavier has written and he told me hey, have your has written mastering spark with our learning spark and engineering production great apps. I only written one book. The first one the other two books. Are incorrect. I have not written those books and this is what is called hallucinations and it would be the equivalent as forcing someone like hey. You know, I wrote three books which which are the three books that I wrote, and maybe they only remember 1 and or, you know, and they they make up the. Three, this could be good in some cases where you ask such GPTS like come up with a poem or a story or the like. It's a very creative answer, but when you're looking for factual answers, chat DPT. And many of the large language models have this challenge that some answers might not be factual. Another issue is you know I'm asking a simple question to ChatGPT are apples better than the? And obviously there's no good answer there. I don't know which food is better, and the answer is telling us no, which is fine. I don't care. I'm asking the question another second time. The answer from chat DPT is again no. But if I asked the exact same question a third time, I get the answer yes. And you know, I wouldn't expect this. You know, when you usually when you interact with a computer programme, you know you are expecting deterministic answers, you know, like that's what we. Select the best example is when we do for instance math and the like. If we ask a calculator if we give them a simple problem. It's like, hey, I have this many employees and I pay on average this much. What's the what's the total compensation budget in this case again? Chat if it is giving us the wrong answer and it has the IT has the right structures like the two first numbers are correct and it shows the correct multiplication. The answer is not that far from incorrect, but it happens to be incorrect. So it's kind of weird that now we have computers that are kind of bad at doing computer like things, you know, adding numbers and doing multiplication is one of the things that you would expect a computer programme or a service to be very good at. And the best example is, you know, computers were invented to, you know, to help us in many ways with these repeated tasks, like adding up numbers multiple, doing multiplication and. And it's it's interesting that chat DPT is really not good at math. It's not good at the things that we expect computers to be good. At. So you know, I hear the complaint is like, why can't change be like a normal computer, right? It's just something that is unexpected. Now, the good news is that that's the last of the caveats that we'll cover. And again, the, the. Point is not to be negative net net charge PT is super positive. You need to be aware of all these issues you know and you know, be a sceptic when you ask chatpad to do a math problem is like you know, I know that you're good at reasoning, but like I need to double check it right? Like that's one of the things. So I don't know if there's any fans of Westworld that's an HBO show, you know about robots and the like, and hopefully I'm not spoiling too much, but there is one of the characters, Bernard. He's he's the director of the research facility. And then at some point they, you know, Bernard realises that, you know, he's not. A person, but he's an actual robot and that's, you know, super interesting. Because I think you know, to me this makes me think that. You should start thinking of ChatGPT less of a computer you know, less of a robot and more of a person. And I don't mean I don't mean to describe charged PT as a person, as a you know it it has consciousness or or the like. I just mean it that. If you try to think of chatty pity as a person, it will help you understand the limitations and also the strengths of the technology. So yeah, why don't you illustrate that point? OK, so now we're we're at the part part of the road map where you know if you adopt chat between your organisation, you know like maybe everyone is using it, maybe everyone now understands like the things that chat DPT is bad at, right and you know like one way of solving the problem is just to say, OK, I know that charge PT is bad at those things you know. I'll just be cautious right when I ask those types of questions. I know that is biassed. I know that X&Y, but our goal again is to, you know, at the end of this webinar to be very much experts to some degree on how to use ChatGPT efficiently so. If we combine the this slide with the previous slide, right, we have we should think of chart GPT as a person for practical purposes, not for philosophical purposes. And then the question is how do I make a person better to be able to solve problems that they're not good at or that they need to grow into solving them? And the analogy here is OK how do we make chatt's a better employee? How do we make it kind of like a Jedi? And if you look at Star Wars plot lines and storylines? Like. There's really only two ways of from my point of view from, you know, helping someone become a Jedi. So when Yoda is helping look become a Jedi, there's there's really two big concepts that you can think of on how to make someone better. Right. So the first one is you mentor them. So Yoda spends a lot of time telling Luke.
Like.
This is what it means to become a Jedi. This is the things that you do. These are the things that you don't do and this concept is gonna be very important for you to understand when. When you're dealing with chat DPTS like OK like I need to actually teach it, I shouldn't just expect answers in all the cases. I need to kind of like, teach it. And the second way is you give someone a lightsaber. Obviously, chat DPT doesn't need a lightsaber. But the concept here is to make someone better. You need to give them tools. So like you literally give them tools that make them more proficient than they like. So those are the two kind of like categories that we're going to discuss to overcome all the challenges that I mentioned and yeah, so let's just get that going. So. So for the first problem that we mentioned, where your data could potentially be leaked to third parties cuz by default it improves the model for everyone. The fix is very very simple. All you have to do is you want to go to the settings and there's a setting under data controls improve the data. The model for everyone. You can turn it from the default from on to off that you turn it off. There's also like an option to create a private chat so you can when you click create you can kind of like an Incognito window when you browse the Internet on private mode chat. DPT has something similar where you can upload data without them necessarily leaking it. Or improving the model with that one. But yeah, the most important thing is you're aware of, you know, how to overcome it. That's one. Now the second one, we were asking the question what are the, what are the, what are my e-mail, what are the emails of my customers? If you think about it, chat DPT doesn't have access to any of your private information. Now in this particular case, if you happen to be using something like hotspot for managing your customers. Or if you're happening to be doing project planning in Microsoft Project right or Scrumming Trail or the like, kind of like what you can look into. These are there tools out there that connect ChatGPT with this data source right? In the case of hotspot, they have something called chat spot that is now part of their product, so you should be able to go to hotspot and ask the question what are the emails of my company. And now since there is a tool that connects them, the two of them together, you should answer those questions. But the same applies to hey, I want to ask questions about my QuickBooks accounting or trial or my GitHub issues or the like. Well, find that tool that connects chat DPT with that thing and that's how you make chat DPT more powerful. Next problem, remember, we were asking if watering the garden is good or not, and we know that JPT is biassed is biassed cuz Chad DPT was trained on all the data on the Internet, and a little bit more so. Whatever biases the Internet has, chat DPT brings them along with them, and also the open AI team made it better. Or improve it so it is biassed right, so the way to overcome this is you can give it additional context, so I could say something like hey, I only care about the things that benefit society economically. Is it good or bad to have well water? And now the answer is telling me is like it's not. It's not good, right? It's OK to not water, water your garden. Basically what we're doing here is we're teaching chat DPT. What are our values or what do we care about? So usually uploading data to chat DPT that is relevant to the question or just copy pasting. Like meaningful information that you care about that also can help judge app like be less biassed, and answer questions closer to the types of things that you care about. Here's one that I got the screenshot from Hal 9 cuz we have a free version but chat DPT. If you pay the premium plan, you can upload also CSV files which is excel sheets and the like. So for the problem of how many books has had here written what you can do is upload to chat DPT premium. Or use a tool like Hal 9 or Claude. We'll talk a little bit more about tools you upload your data right? Like, hey, these are my, this is my Excel spreadsheet with all my the books that I've written or all my customers or all my revenue or the like. And then you ask the question is like, hey, how many books has have you written? And now the answer is like, it's only 1. You know, it's mastering spark with our from O'Reilly. And again you can you can do this with multiple tools. I just borrowed the example with online. This one is actually a tricky one. The one that is random just the way that I works today there's. Randomization in place, so asking even if you ask one day change PPT one question and you ask the next day the same question, things are gonna change, but you can make it. You can reduce the. That. Randomness, right? By providing additional context and information to ChatGPT. So in this case for the apples and bananas question, which one is better? I I told Chad PT's like hey, I I care a lot about, you know, foods with more calories are better. You know, it's just, you know, it's the additional context we're training charge PPT with the. Assumptions that we have with the premises that we have and then we're asking the question three times, if it is still random, but it's using the information that we're providing and the outcome is less random. So there's active research going on here. But what I'm saying is the more information that you provide to ChatGPT that is relevant. The less random it will behave, and I think that's that's good net net. The example about you know the the particular problem that we have on doing some math, same answer than getting ChatGPT premium. So you can get chat DPT premium which comes with a feature called Code interpreter which basically it's a calculator that comes attached to chat DPT. Premium. So you can either pay for chat DPT premium and then you get access to giving chat DPT. A calculator which is good in our platform. How 9 we have a free plan and you can add those types of questions and we give you. We also enable chat DPT with a calculator and again the tool matters less. What is important to know here is that. You want to you know, if you're gonna be asking math questions. And you you want to rely on the answers, you should probably pay for chat DPT premium, or use some other tool that has access to that gives chat DPT access to a calculator. Now the. You know, this covers, you know, kind of like this, this section of improvements, you know, on ChatGPT. But you know, you might think. Oh my God. You know, kind of like Tony Stark here is thinking this looks like a lot of work. Like, I basically need to train all my employees, my team. I need to tell them how to use charge PT. You know, like I need to help them a lot. This. That sounds like a lot of work. Is there a better way? And the answer is yes. And we're getting. We're getting to the point where we're kind of like on the most. And you know, usage of chat DPT. So the good news is that we're there. So you made it all the way here and we're gonna look at some of the examples that can help us automate the tasks to help our make our team more productive, right and not have to go through all these learnings. So again, these are examples that we've done in our company, so there are truthful examples that save us a lot of time. So one of the one of the integrations that we did is. We didn't want to spend our daily scrums, you know, like, you know, with a recap of what everyone did on the past day. We wanted to spend time unblocking people and discussing important things. So what we did is we we used CLOCKY, which is a tool to. Track our tasks in our company. They have a free plan so everyone that whatever we work on, we put it on clocky and we do that for invoicing and the like. And then what we did is we connected ChatGPT with quantify. OK. And ask chat DPT hey every day at 6:00 AM in the morning. You know send us a message in the slack that shows who work on what. What is the summary of the tasks that were accomplished the previous day? How many hours did we put and then the breakdown of what everyone worked on? Now what this enables us is. There's going to be times where we don't need to meet for Scrum, and when we do meet for Scrum, we can spend those 30 minutes actually talking about the blocking issues versus you know, like giving a recollection of what everyone did the previous day, which saves us time so. We've been doing this for two years and a half. Unfortunately, this is like a custom little tool that we built, you know, just for us. But but the learning here and what I'm trying to explain is. You know, like the Jedi level of using chart, GPT is to think, OK, I understand chart GPT. I understand their problems. I understand how to overcome them. But now I have a team dynamic right? Or I have a particular problem at work where I need to do all of that automatically and and the answer is yes, it's possible, but it's dependent on your use case and you know, I do believe that each company is slightly different, right? Maybe taking notes is not the issue, but reminding them to do their work throughout the day. Maybe that's the challenging thing, right? And you want to send reminders? It's like, hey, you were going to complete this task and the like. So, I mean, I wouldn't spend too much time here, but hopefully this. Gives you some idea of like the types of. You know things that we improve in and I'm gonna try to go fast because I we have so many slides, but I do want to cover also the questions that you have. I'll send the slides and the like. So I'm going to switch gears and go a little bit faster. Another use case, we use GPT to embed it in our our website. I think that's a good use case. To answer questions about your website for the startup. It didn't work very well because we don't have a tonne of customers that come to our website anyways and we want to talk to you, but it might be the case. It might be a good use case to automate customer support calls with chat DPT. Another one you know, we have an integration of using Char GPT directly in Slack that has some value. Save us a little bit of time, but it's middle middle of the of the spectrum I guess. And this one I really, really do love. We generated a lot of our content in our website for the demos and the industry verticals. We automatically generated all that content with chat DPT. You don't see it when you come to our website, you go industries, you know demos and the like and you see images. But all of this was automatically generated and this probably saved us a few weeks of work. Yeah, so that that was a tonne of savings and, you know, real use case that we did. Now we could have done it by copy pasting to chat DPT one by one. That wouldn't be that bad. It would have saved us instead of two weeks maybe one week. But yeah, we basically automated that process. We generated all the content and and then we were done. And this is just a quick learning. I I think these types of automation projects and chat DPT are tricky and the best way to tackle them is incrementally. So divide and conquer. It's a practise that I'm sure many of you are familiar. My my suggestion is if you go and try to do an automation project with chat DPT. Don't don't make it huge in scope, but break it into very small pieces and try to get a quick win. It's like, hey, let's just, you know, like post the messages from. Clock defying to slack. That's D1 and do a summary. OK, done. And then you get the value and then you iterate and mostly cause it's still a new technology and it's hard to get everything right on the first try. Alright, I'll just use that 5-3 minutes on tools that we use and then we'll I'll switch to the questions that you have. So tools that we have explored and this is related with your original question at the beginning of this webinar. We've looked into probably hundreds. Not more than 100 tools that all do AI. There is tools for as one of the participants was mentioning, right, like taking notes like other AI and read a I can take notes of this presentation. Slack has an integration that allows you to use ChatGPT in Slack for an extra $10. A month. That's another tool. There is cloud, which is better in some degrees than ChatGPT. And again, we're taking a look at that. If you're writing code, there is a warp terminal and there's compiler that can help you write better code. And you know, to be honest, I think this would be an entire webinar, you know, presenting hundreds of tools in the industry and, you know, finding and the like, there's, there's a lot of them. What we have found out. And again, this is my our own company. So it doesn't need to be necessarily the case for yours, but really the tool that we're using the most. These ChatGPT. So there there's three tools that we really use, but 80% of the of the work goes on chat DPT and the reasoning is that. Rather than using the summaries from other AI, you know just to use build on that question from a participant. What we found out that has the most value is moving all the data to chat DPT. Why? Well other AI is going to give you a summary of this webinar, but it might. It might focus on the wrong things right? It's like. Maybe just three paragraphs that is very broad overview what you can do is you use heavily chat DPT's like you can take the transcription from other AI. I think that's really useful. You can paste it into ChatGPT and then say hey, you know, like I really care about the part where Javier was exploring tools. Can you make me a summary of just that area right or can you elaborate more, can you add additional points? So what we found out is that we're the most productive when we use charge DPT heavily and push more. Data into ChatGPT. Same for coding. I use chart GPT for coding a lot, but I don't use an IDE with ChatGPT embedded. I I use chat app for everything as in hey you know I want to build a website for how 9000. What colour should I use? Orange. OK, that's great. Drive me the skeleton of that website. OK, that's the thing I have. Now I move it to my ID and. The like so. We use AI a lot in our company, but we disproportionately. Use chat DPT and we use very other view. Few tools, the other tool that we use is called mid journey and flux and again this is this probably deserves an entire webinar on its own, but you can generate images with mid journey and flux. I'll show you a few examples. And the third tool that we use, and again I'm biassed, so don't don't don't believe me on this, but like we use a lot of the company, the tool that we built on our company, which helps us build automation. So it's kind of like for the people that are at the very end of hey, I want to automate X or Y. So that's what we use online for. But yeah, if you wanna try flocks, we have a free version. And this is literally the last few slides and I'll get to. You can go to hotline.com apps, flocks or just click hotline and then look for the flux model and here are some images that we generated just for fun, like baby Yoda using an iPhone in a coffee shop, in selling lemonade, climbing a wall and doing pancakes. So just one of those things that. We use this a lot for marketing it just for fun as well and the like and feel free to use. Yeah, that that app that we generated and look at it. All right, so. We have, I don't know. We need to hopefully finish on time. We have 9 minutes and I have a few questions that are that we got submitted before the webinar. So the first question is like how can you future prove your role in QA test against AI? And again, I think the questions really apply to all the stuff that I said. In this webinar so like you can apply the concepts to answer all these questions. But I wanted to give like a real use case on how would I would I would improve my QA testing with JPT. So this is a real piece of code that I got from how 9 it helps us do authentication and it's one of the things that I would want to queue a test, right? It's like hey, I should probably do this better. So what we did, we chat. It is. I copy pasted the code and again this follows the pattern of, you know, being a very smart co-worker like help me solve a hard problem and I was like, hey, what things should I test on this function right or on this feature? And I got five different, you know, different Qi testing things that I should consider. And again, you know, like many of them, you probably already know them, but maybe maybe there's there's a blind spot blind spot that you have, maybe something that you didn't spot it. So using chat, DPT for QA. In this way you know it's it's a great way. Now the more complex 1 would be. You can also build an automation right? So you can look at the API that you're testing this is using. I just want to call postman and then you could potentially use a tool like chat DPT or Hal 9. Build me an application to test this URL and you know like you generate the code to test that particular application that's more on the automation side, but that's kind of like the type of use cases that you know, like the future proofing basically means you need to embrace generative AI in your workflows. You know again like you have two examples. Money is, you know, getting brainstorming with chat DPT and you know like getting your testing coverage broader and just by the nature of using it. And the second one is trying to automate some of those processes and the. Like so, there's another question is what advice do you have for non-technical it roles. To keep unemployment advantage with regard with AI training and related courses. I don't know if I fully understand the question, but what I wanted to share here is I wanted to find kind of like the best course or you know like using. Think. Yeah, I am the like for training. I think it's still very early on, it's it's kind of like tricky to find that information. But I think my best recommendation is two ways. One is you can follow our company on LinkedIn and we try to do posting of like new things that we're following and the like. The other one is if you search LinkedIn for ChatGPT with a hashtag, it should take you to a page that looks like this. So you can like it says charge PPT has 100,000 followers and then you can follow the hashtag and then basically you're gonna see on LinkedIn. Ways that people use chat. GPG is like hey, you know, like there's this new app, there's this research and the like. It's not perfect, but it's like a way of keeping up to date with chat DPT. Right now it's very early on. So that's the good news. There's still a huge advantage to be taken, but it's also kind of messy. And I think I think I have like less. Particular courses to recommend, but rather be part of the community and you know like be some of the people that are innovating in this space and the like. Yeah. Another question somewhat related. As a programme manager, how do I best recommend with direct product and automation teams to lice? AI, perhaps standardise AI training framework? Is the pipeline at an international level? Any thoughts? This question the answer is like it is like a really hard question but what I want to say is. If you want to talk with me and our team, you know, feel free to book us so you can go to hotline.com and we can talk about your particular. Answer you know issues. We do have a training for AI. The challenge is that we have a training to use our tool, but I can try to find out a partner that could potentially do the training. So the short answer is I don't know, but if you want to book me 30 minutes, we can talk about the particulars and I can point you in the right direction. At our company, we don't do training per sale of ChatGPT. I do have a couple of pointers that maybe I can recommend you to as long as you have some budget and the like. But yeah, it's like a tricky 1. I didn't know the answer on top of my head and but happy to chat. That. This one is, I think is super interesting as someone has a question. What particular task of HR can be managed with the help of AI which would which I would be the most suitable for this? What information is safe to upload to AI? Our personal records safe to upload to AI? So again probably you should book me and we should talk more about this. The short answer is like yes there are solutions. For everything you can run AI privately in your cloud. I teach you how to turn off the model. You can also use a private model and the answer of which HR tasks can be managed, which AI. The answer is all of them. But the bad news is that there's no tools yet that automate yet all all of these for you, or that I know. So let me you know what I wanted to show show is, and I'll. I can send the link or you can find it online. But there's a company called Lemonade. They're an investor. They're like an insurance company. They just did an investor. A presentation day where they show how they use AI on the back end. You know, like on on their HR processes and the like. And then they had to build their own integration. So it's kind of like being a Jedi of AI automation, but here are some of the examples so. You know, first of all, they're trying to really embrace AI in their DNA. And I borrowed these slides, hopefully that that's OK with them. I mean it's it's public and they have a lot of their processes on the back end. They, they, they use the. I totally made the tools. So here's one example, for instance is I think this one was for procurement. So when there's when there's an expense. That goes above the limit, you know, like the AI will send a message and say, hey, you know, like, can you approve this? Yes or no? Right. I think you should approve it. But is this OK? Yes or no. And they build like a custom integration to do. That right, they also are doing performance reviews with AI, so they built another automation in Slack that does the 360 feedback review now. They do the collection automatically and then they build the summary. Now a manager still needs to review the 360 review, but they basically built a kind of like a little automation that automatically goes and asks feedback for everyone for every. 1. Yeah, everyone for every all of their colleagues that need the 360 review, they, they they save it internally and then they build this report, right? I was like, hey, you know, like how are you performing against, you know, like your peers and you know like is your is it going up or down and and again I think this is like a great way of using AI and automation in the. In the workplace. But the bad news is what I've seen is that this is so early stage that a lot of the automation that I've seen are kind of like. They're ad hoc tools that are getting built for organisations and less of products that are already ready to use, but you could also search online and there might be some of them and the like. But yeah, I mean, maybe we have time for one question up to you. You wanna? But if you want to connect with me, you can find us on online.com. We have a clock. Link there to chat on the like and happy to spend a little bit more time if one of you have a particular question or problem that you want to explore, yeah.
So Javier, we actually have one more question on the chart. So when you were talking about the pitfalls and setbacks of that PT, the question was written down. Do you end up writing really long prompts in your day-to-day to avoid the mentioned caveats?
Yeah, so, So what I do is I use the history. So the answer is no and yes, I start with very short prompts like I try to always. It's like, hey, you know, like which, you know, should I should I use Google Cloud or Microsoft right. Like, that can be my prompt and then I get an answer and then I interact with chat DPT and if the answer feels like they're not, it's not considered X or Yi copy information about our usage of cloud compute, right? So. The answer is yes at the end because I end up iterating with chart GPT. You know over a few a few ways back and forth, but at the end of the day I do end up, you know, the prominence of being long in the sense. That I iterate which adpt a few a few in a few different ways and I pass the information that I know that charge if it is missing to answer the questions correctly. So yes, but I don't do it proactively, I do it more like reactive. It's like I asked the question. The answer doesn't feel necessarily correct and then I add the additional information. Sometimes I also come back to history. Previous previous history. So I would you know if I already asked a question about should we go for cloud for Google Cloud or Microsoft, that's our cloud provider. I opened the history where I talked with how 9 sorry with how 9 or chat DPT about that it doesn't matter. And then I actually asked the question he's like, OK, like I'm gonna ask a follow up questions. So I use the history. And I do a lot of copy pasting. That's kind of like how I use it but but yeah, that's that's the the workflow that I have.
Thank you very much for sharing. If you have any more questions please. Either ping have your himself via her, mine had mine or you can ping us. You can send message to us and we will be very happy to follow up. We have here regarding that.