Episode transcript The transcript is generated automatically by Podscribe, Sonix, Otter and other electronic transcription services.
Hello everyone. Welcome to the 5 Minutes Podcast. Today I like to go back to generative AI, and I want to talk about people coming to me and saying, Ricardo, you know, I'm using ChatGPT, I'm using Bard, I'm using Bing. And you know the answers I got, they are not good. Some of them are wrong, and some of them are incomplete. One of my friends translated this into a very nice concept. He said it's pasteurized information. Do you know what Pasteurized information is? Pasteurized information it's, you know, information that is full of words but lacks a lot of meaning. And then they said, you know, I cannot rely on this. And then, I want to share with you my rule and the rule I use in my daily life. I share this on my latest YouTube video about the PMOtto and also on my new LinkedIn course, Generative AI in Project Management, which was just released by Linkedin Globally. And on this specific lecture, I talk about the rule that I call 5 to 80, or you can talk about 75 to 85. What do I mean by that? It's 75% of the work will be done by ChatGPT, Bard, Bing, or even by PMOtto. But 25% it's on you. It's on your human brain. And you cannot delegate that to any AI so far. And this brings me to the 5 to 80. What is 5 to 80 First, The first 5% of your journey through your answer belongs to you. It's to make a clear, nice, and sharp prompt. If you put a poor prompt in your AI, in your ChatGPT or in your Bard, the answer will be a reflection of that poor input.
The chances of being very poor are very big, so you need to build a prompt that is CSIF. C, It provides context. For example, if I ask ChatGPT to say what is the average cost of construction of a house, you know, it may come with an answer. I haven't tried, okay, but it may come with an answer. But what is the reality of that answer? But if you give a context, for example, I am preparing a marketing study to release, for example, a new incorporation of homes in that specific city. I'm planning to build homes with three bedrooms, whatever, and I present the characteristics of this. I'm being specific. I'm giving context and being specific, and specific is the S of CSIF. The third one, you need to explain your intent. What do you want ChatGPT, Bard, and Bing to do? What do you want them to do? You may say my intent is to have the list of average prices of this kind of house in this kind of city in the past ten years, for example. And the F stands for format. How do you want your output to be? Do you want a list? Do you want a table? Or for example, if you are using a ChatGPT code interpreter, you can even ask for a chart. So did you see that? Look, I'm making up this example, okay? Don't try to use this example on ChatGPT because I don't know what will answer. But what is important is that the quality of what you insert will be the quality of your output.
So this 5%, you cannot be, I would say, lazy about doing it because the chances you will receive an answer that will not be the best one is very big. Just to give you an example, when we see these wonderful pictures of AI-generated pictures that look so real that you don't win a photography contest using fake art or this, Do you think that these people did one prompt? No, not at all. Sometimes to build that quality of image required 200 or 300 prompts. It's an artwork to make that happen. It's not just saying, Oh, create an image of a house with the sun in the back and this, and bingo, it comes perfect. It's not true. This is not true. So you need to understand the quality of your prompt. Now, from 5 to 75, you will let the engine work. You have the algorithm, the machine learning, building these neural networks to get you the answer. And then what you should never do is copy that answer and paste. Where you want it to be, this is the path to make something really bad. I remember people telling that one lawyer took an answer from ChatGPT and put in a petition to the judge, and it was wrong because the law didn't even exist. And he was mad with ChatGPT. No, the word needs to be mad on him because, you know, every time ChatGPT, Bingo, or any other AI-generated text or image comes out, you need to spend that 25% analyzing the results, tweaking the results, making sure the results are compatible with what you want.
I use ChatGPT. I use Bard all the time. Every time I'm extremely cautious about making sure that I'm reading. For example, if it's a list of risks that I'm reading that that makes sense because on the top of seven risks, that makes sense may come one that does not make any sense at all. What all AI companies are working on now, it's how they can reduce the 5% and reduce the last 20%. It's something like, instead of you giving 5% on the prompt, can you give 2%? Can you give 1%? And instead of bringing you up to 80%, and you do, the 20% that is remaining is to deliver something that is close to perfect, 95%. This is exactly what everybody is working on. And maybe in the future, it will deliver a perfect answer. And personally, I think it will happen, but it will not happen now if you use ChatGPT tomorrow. Be just mindful of that because many times I see posts on Linkedin that I know were produced on ChatGPT, and the author did not even have the discipline to read it because, you know, in the middle is just pasteurization of language and with no meaning whatsoever. So always think about that. Think about how you can craft your best prompt and analyze the outcome on the other side. This is an intelligent use of artificial intelligence. I hope you enjoyed this podcast, and see you next week with another 5 Minutes Podcast.