 Hello, everybody, and welcome to our sixth AgriTech Talk. An initiative of the Regional Office for Europe and Central Asia led to inspire us all to think tag when designing and implementing our projects and programs across the region. My name is Veronica Sherva, and I am the Digital Agriculture Analyst and today's host. Get ready to dive into the exciting world of artificial intelligence and child LGBT because today's AgriTech Talk is for us to really try to understand whether child LGBT and technologies alike are going to revolutionize our work and agriculture. But before we start, let me quickly ask you because I'm really curious. Do you use child LGBT in your work and for which purposes? Please answer in the open survey window and be mindful that the results will be kept anonymous. Our special guest today is Padron Marcello Ayes, economist at RAO, who will tell us about how to use child LGBT, what potential in agriculture it has and what are its limitations. Padron, welcome and thanks for joining us. Thank you, Veronica, and good morning or afternoon to everyone. Thank you, Padron. So why is everyone talking about child LGBT? Well, the short, yes? Just let me tell quickly that we'll have a timer, okay, so I'm gonna start the timer now for 12 minutes. Okay, so we need to be very timely, so that's why we're using it and we'll have 12 minutes and four questions to be answered in this time, okay? Let me start it quickly and ask my first question. Okay, you should be seeing it now. So, Padron, can you give us a little background on child LGBT and why is everybody talking about it? Yeah, yeah, sure. I'll keep an eye on the watch, I'll do my best. Yeah, well, the short answer is because it's just amazing. I mean, this is a program that passed an MBA, a medical degree, even degrees in law. I wrote somewhere in Teams, actually, that if you're not interested, it's only because you haven't tried it. So, chat GPT, generative pre-trained transformer. I've been looking at this and the difficult part for me to understand is the transformer bit. I think I understand the generative and pre-trained part of it and that helps me to understand why it does what it does and what it could do to me. So, chat GPT is what is called a large language model. It was developed by OpenAI, which is an NGO. And so, what is a language model? Well, in simple words, it's just a mathematical model of language. Instead of words, it uses numbers. It's a language of numbers, so to speak. So, if I say, for example, food in English, alimento en español, pasto in Italian, these are all different words and yet they mean the same. I don't know how you say food in Hungarian, I apologize, or in Russian. But anyway, the idea is, why can't we use numbers instead? So, well, actually what the program does, it doesn't transform language into numbers. They describe these as tokens and the numbers are not actually numbers, it's what is called vectors. But those are technicalities and I'm not going into it. I am a user, so don't ask me. So, chat GPT is then this model that was trained into a massive amount of data, books, articles, websites. It actually trained itself. Yeah, it gets this cloud of data, analyzes it, assigns numbers, finds association between these numbers or vectors. And it does this alone. This is the amazing thing. It's its own teacher. This is what's called machine learning. The problem is, and somebody had been talking about this, is just nobody knows how much it knows. So, some people are getting scared about these artificial intelligence things. So, how does it work? I mean, you ask a question or a prompt, what's called a prompt, and chat GPT generates texts. For example, on food security. It locks this concept, this vector in space and starts to associate words that are related to food security. The transformative of GPT, the transformative is actually what it does, it signifies that it actually focuses on something. So, on the question that you put, you focus, you ask it to focus. So, the question is very important. This comes from a paper that was developed in 2017, very recently. It's called attention is all you need. I read it, I didn't understand half of it. Anyway, I'm amazed. So, I believe that chat GPT is popular because it's just amazing. Anyone who uses it is amazed, including the developers, they say the developers, they does more than they ever imagined. So, back to you. Thank you, Pedro. Very useful to receive such a comprehensive background. And I see from our poll results that actually majority of our participants today have not used chat GPT. So, you know, why don't we show them how to use it? Would you be able to do that? Sure, let me just accommodate my desk. Let's do that. Give it a chair. Let's open it. Let's see, I need to share my screen, right? Share the screen. Can you see it? Yes, we can. OK, so this is what you get when you subscribe to it. You know all you need, I think it's just a telephone number where they send you a code, it's free. They ask people to work on it. And because they are testing it and the information here is feeding other language models. So, this is this interest from them to do this. So, let's log in. Let's see if we are lucky. Oops, let's try again. Continue. OK, we are in. So, I'll explain to you how I use this. Since this has all the text that has been fed into the Internet, including that, the FAO actually has read this FAO strategic framework. That's to give you an idea. So, you can think in terms of the strategic framework. I use this as a research tool. As a research tool. One of my difficulties, and I'm sure this is for everyone, is when you're preparing a project and you're doing the log frame, you have between the log frame assumptions that you need to make in the logical framework for the activities that you're going to do. So, let's explore these assumptions. So, let's see. Hello there. Hello. How can I assist you today? So, we started to talk with this thing. With this beast. Can we talk about assumptions of log frames in projects? You don't even need to write it correctly. It has a spell check. So, it understands even the language that you're using. Sure, it says so it goes into a long explanation of what log frames are. And, you know, sometimes it just goes for too long and gives you more text than you asked for. Because this is what it has been generated. It's been, its objective is to generate text on this particular issue. Now, you have to go back here. This is important. Assumptions, log frames and projects. This is where I told you that it's going to look at that space. With this prompt, with this question, I'm already telling it, it's about projects. Within projects is about the log frames and within the log frames is about the assumptions. So, with this question already locked, this conversation, because it remembers what you asked before, I locked this conversation in these three pillars here. Project, log frame assumptions. And we're going to talk about this, right? So, I find it very hard to identify the assumptions of one of my, of the activities of my project. Can you help me? Of course, identifying assumptions can sometimes be challenging. This is just fantastic. I mean, it just throwback to you what you expected, yeah? But it goes on and on again. I mean, this happens always. I mean, it's good in a sense that it gives you all this because that's exactly what I need to think about the assumptions, right? So, the what if risk independence or what, yeah? So, it already starts to generate issues that could be related to assumptions. So, if I say, how about regulatory and policy environment? Would that be assumption for a project, for an activity that aims to integrate the farmers into a global market? So, that's the one I'm worried about. As everybody knows, we want to put small holders in, I'm an economist. I want to put small holder in contact with markets, right? And I want to say, well, if I do this, what am I assuming, yeah? What am I assuming that pretend that this is actually ever going to work, right? Yes, the regulatory and policy environment can indeed be a key assumption for an activity that aims to integrate small holders. And then it goes on to look at all the, if not all the most probable issues that surround that discourse that is being going around on integrating small holders into action, into markets. I could even go deeper in here. I mean, I'm sure you're familiar with this. It complies with social and environmental standards, supportive institution and framework, stable and fabricated trade policies, complies with export and import regulations. This is exactly the kind of issues that we look into the FAO. So it's just, I mean, it's mind boggling what this thing is doing, right? It's gone through that space of text which I asked it to look and starts to spell out all the issues that go around it. Of course, then I may be interested to say, okay, I can see that my project is highly risky to the policy environment because the policy environment is not secure. So I say, what could I do to mitigate the risks? And this conversation can go on and on and on. And to me, it helps me very much. I mean, it's incredible. It helps me to get my ideas through to understand issues and concepts in very, very depth, very, very, very in depth. Well, yeah, I'm sure GTP will tell me the correct language and have to express this. Back to you, Veronica. Thank you, Pedro. I'm very interesting. So you basically use it as a way to bounce back your ideas, you know, to reflect on some of your ideas and so on. So what do you think are other potential use cases? And, you know, importantly, what are the limitations of chatGPT that we should take in account? Right, that's a good question. I think that it's important to understand the basic features of chatGPT so that you understand what you can do in the potentials, yeah? This chatGPT was trained in data until 2021. So it doesn't have anything from 2022 or 2023. The other thing is that whatever you write in chatGPT is stored by the NGO, by OpenII, for the purpose of training future generations of GPT, yeah? So you have to be careful what you put in it. Our colleagues in Rome don't have access to GPT now. It was locked because of that. Because whatever information Italians put is going to be stored somewhere, you know, sort of privacy laws and that kind of thing. So another thing that we need to understand is a large language model. So it's sole objective in life is to generate text based on a prompt. It's larger than life. So if it doesn't understand something, it bullshits. It generates text, generates data. It doesn't know something, it just tells you anything. So you have to be very careful. And you have to know your subject. And that's how I use it to research. I know what I'm talking about and I know what is bullshitting me, right? Excuse my French. So it's a language model. It doesn't do math or logic. Though sometimes you wonder, you know? Katrina actually was sharing a video that researchers are beginning to think that actually it does, it does a lot of reasoning already. Not chatGPT, but the next one with GPT4, which is there. And as I said, I mean, the secret of using it properly is in the question, in the prompt. You want a good answer, you ask the right question. If you don't know how to ask, if you don't know the prompt, then ask GPT itself. Could you please help me to formulate my question? I mean, just take a minute to think about this again. You ask GPT to help you to formulate the question for what you want to get. So chatGPT can be used in many ways. It's difficult for me to explain. I think the only way really is by using it. You use it and you will realize then what can be useful. And again, I'm sorry about our colleagues in Rome, but there's just so many language models out there that are going to be available. I think it's only a matter of time before they give it access again. Back to you. Thanks, Pedro. Indeed, we shared in the chat some link to discover other language models. So dear participants, feel free to explore. So we ran a little bit out of time. So let me ask just our last question, which is, is chatGPT being used today in any application, including for agriculture? And do you think it will disrupt the sector and now work in the future? Yeah, it's a very good question. Because it's just a language model. And yes, it's been already used with our applications. Ball from Alpha, for example. One of the difficulties is it doesn't compute, right? So if you add this language, which asks another model to do computations for it and gives you back the answer, that it can do, it has that potential. And in agriculture at the moment, I think it's just for research purposes. But I can see a future where you ask a text a question, it's linked to, for example, I don't know, let's say remote sensing, yeah? Remote sensing where you have a map and then you ask, tell me about La Niña, it's over now, La Niña is coming up. What's going to be the prospects for Kazakhstan this winter? And it will go into remote sensing, knows how to analyze it, sends you back. You don't have to be researcher, you just have a language, a dialogue with this text without having the need to go deep into understanding the issues. But as I said, you need to be an expert really to make the most of it. So yes, it has potentials, it's good, but it's also, it's bad if you don't know how to use it or what it can do. I use it for research, that's what I do. I hear some people are using it for editorial purposes, translation, to write speeches, computer programs. So yeah, many uses. Back to you. Thanks, thanks Pedro. So let me see if there are any questions from the audience. Aida, I have been monitoring chat. Are there any questions, concerns? Thanks Pedro. Anything about participants? Thanks Pedro, thanks Veronica. Thank you for these informative responses Pedro. Actually, we have a few questions in the chat and I will read out one of them due to time constraints. So there is one that says, does it ever, does this, meaning chat to PT ever give a nonsense answers? If yes, so that means that it must not be very reliable. That's a good question. I don't trust it. I personally don't trust it, you know? As I said, it helps me to think through the issues. That's what it does. It's not going to do my work. It's not going to, I mean, the fear out there that it's going to replace workers, knowledge workers, not yet, maybe in the future. I use it as a bounce ideas to get ideas, to explore issues. And as I said, if it doesn't know something it bullshits, so you have to be careful. I personally don't trust it. Okay, so then we must take mitigating measures that will check the information, right? And always adhere to some trusted sources. Yes, of course. I mean, you have to be, know your subject, you know? It's very imaginative, you know? So it can go into an explore territories, which is good in a way because issues about parallel thinking and so on. So yeah, I think it's, I mean, to me, as I say, I don't trust everything it tells me, no, but it's very good to talk to, to explore an issue for sure. Okay, well, I definitely agree with you. Thank you very much for joining us today. It was really great to learn from you about this application, its potential and also its limitations. We maybe did in just a little bit about 20 minutes. So thank you very much. Sure. I hope that people get the interest in GPT. I think that you should use it and it's coming. I mean, GPT for trust GPT, Elon Musk is coming up. There is the other one by Google namescapes. Bard is coming up as well, right? Bard, yeah, which is in the United States and in the UK, I think it's not released by, for other countries. And there are just hundreds. It's going to be a whole ecosystem of these things. And FAO will probably jump into it by creating its own GPT with all the basic texts and everything that has been published by the FAO that people working in the FAO can refer to, right? I can see that happening soon and it should be done probably. That would be very interesting. That's actually a very good idea. Yeah. We can take it from here, from this conversation. Okay. Thank you for inviting me. Thank you, Pedro. Thank you.