Become a prompt programming pro: Using ChatGPT and generative AI for everyone - ChatGPT Tutorial

- Now, I've emphasized that this is going to change the way that we interact with computing, that generative AI is going to make us able to control computing using a better interface, in my opinion, than the GUI or programming, that it's going to give us something unique, new, much more flexible and much more intuitive in many cases. But I want to give you a sense of how it's also going to change the types of things that we can do. We think of programming as something that involves going and putting in source code, like, you know, creating source code and programming in Python or some programming language. And even in the examples that I've shown you, I've shown the tool going and generating code. We actually don't even have to have code anymore to begin programming, and I want to show you what I mean by this. Everyone can program with prompts, and I want to show you what this means. So I'm going to go in, I'm going to create a basic program. And here's my program. I'm going to say, whenever you generate output, turn it into a comma-separated value list. So a comma-separated value list is basically a format for describing tabular data like you would put into Excel. It's just basically a series of columns separated by commas. So like name, comma, roll, comma, whatever you want. Jules White, comma, teacher, that type of thing. So I'm going to say, write a program which is basically, whenever you generate output, turn it into a comma-separated value list. Now, notice, this is just text. Now I'm going to go and run my program. I follow up in the next prompt in the conversation and I say, my name is Jules White and I'm teaching a course on prompt engineering. And that's the input to my program. And what does generative AI do, or in this case, ChatGPT? It says, great, here's a CSV list with that information. Name, comma, course, Jules White, comma, prompt engineering. That is CSV, I've basically written a program that converts whatever I put in the next prompt into a CSV that I could copy and paste into a file, save, and import into Excel. Now, notice what I didn't do. I didn't write, you know, a bunch of programming language instructions. I didn't go into great detail on how to do this. And in fact, if I had a program, this would probably take a lot of work to do the natural language processing. And it figured out automatically how to map into the columns. But notice I didn't even tell it what the columns were. I didn't tell it name and course, I mentioned them when we were talking, but I didn't put it into my prompt. It just figured it out, that it needed to generate something that worked. And this is one of the ways that generative AI is really different with computing, is it tries to come up with a solution that works. It doesn't require every last detail to be specified. It's constantly trying to generate something that will work. And so it makes it more accessible and user-friendly. So it generated name, comma, course, it then appropriately semantically mapped, and notice that capability is a really complex, sophisticated capability that would've been really hard to do any other way before these tools came out. Now, let's say I don't want just name and role. Let's say I want more than that. I can simply go and modify my program, and I follow up in the conversation and I say, from now on, the columns of the comma-separated value list should be name, comma, course, comma, role. So I'm modifying my program just by telling it, hey, here's the new rule. And then I go back and I give it exactly the same input in the next prompt. My name is Jules White and I'm teaching a course on prompt engineering. And it says, here is your information formatted as a CSV list. And now it says name, comma, course, comma, role, Jules White, comma, prompt engineering, comma, teacher. And notice what it's done, it's modified my program and then it's run the new program that I specified on the same input. This is exactly the same input I gave last time, but now I'm getting a different output. I have a different program that's running and it's giving me different output. So we can actually go and start programming through prompts and having complex logic just in prompts themselves. So let's say, for example, I want to modify it further, I just follow up and I say, in addition to whatever I type in, generate additional examples that fit the format of the CSV that I've asked you to produce. Now, notice, that's a really, really hard task. And before these tools came out, that would be extremely difficult to generate additional fake data that fit the format that we've just specified. So not only now are we programming, we're also tapping into some of the most sophisticated computing capabilities that we've ever had, and we're doing it on the fly with some text. So all I've done is modified my program by giving it this new instruction. And then I run it again on my original input, which is my name is Jules White and I'm teaching a course on prompt engineering. And now it produces new CSV that follows the new program, which is name, comma, course, comma, role, Jules White, prompt engineering, teacher, and then a bunch of synthetic data. And that is completely transformative in terms of capability. Now let's make it more concrete. Now, CSV, don't get me wrong, like extracting structured CSV from all kinds of data is a hugely important problem that will have really important impacts in all kinds of domains, everything from taking focus groups and turning the conversations in focus groups into CSV that can then be actionable and have concrete product insights to taking, you know, what people are saying and how they're reviewing products and turning it into concrete information about which product are they reviewing and what's their feeling and what are the features they mention, all kinds of different capabilities. Or free text, you imagine giving people surveys, and they enter in all this free text information, and now you've got to laboriously look at it. Well, you can actually turn that free text directly into CSV or some other structured format like JSON for your business to then automate and do things with it. But I'm going to give you something more close to home probably for all of us. And that's that weekend problem we face when we decide to just check out and not answer our email over the weekend. And I don't know if you're like me, but what happens is is I come in on Monday morning and I start looking at my email, and there's some gigantic email thread where everybody else didn't check out over the weekend and they all had a very long conversation that I now need to read and try to figure out who said what and when and what am I supposed to do, what are my action items, what are the questions for me, what do I do about this? So I'm going to use prompt engineering and I'm going to write my own program to solve this problem. So here's what I say. Act is my personal assistant. Whenever I give you a long email thread, you will create a bulleted list of questions and action items for me. The bullets should be numbered. If I type in a number of a bullet, you'll summarize the conversation from the email chain around that bullet in one paragraph. If I type in a bullet and a number, you will draft a reply for me to read with that many paragraphs. So I'm creating a whole little application for myself, and it's even got a menu system, I can type in an identifier of a bullet and it'll summarize it. And if I type in the identifier of a bullet and I give it a number of paragraphs, it'll just draft a reply for me. This is like my email accelerator, right? And all I did to program it was give it a block of text explaining clearly what I wanted. Now I'm going to go and cut and paste a very long email chain when we were scheduling my generative AI class that I'm co-teaching with four other faculty members. So there's five of us on this email chain plus a number of other people in the university. And we were trying to figure out what room and time and what is all the information for this course. Very long email thread, lots and lots of discussions, lots of people weighing in on different aspects. How do I go and decipher this? Well, I'm going to copy and paste that long email thread in right after inputting those instructions, and note, this is one of those important things. Everybody tells you, oh, it was trained on 2021, it can't do anything after that. Of course it can, you just have to give it the information in a prompt. So I'm giving it my email chain in a prompt. And then what does it do? It says, here is a bulleted list of questions and action items based on the email thread. Questions and action items. What are the suggested title and transcript for the course? And then it has the title, you know, possible examples of what I'm looking at. Can we start working on a draft syllabus? And then look at this 1.7 at the end, it says, have the room and schedule been finalized? Tuesday/Thursday 9:30 in ESB 048. And that's actually real information that it's dissected from the conversation. And that would've taken me a long time to go and find all of that, follow up on it, potentially. So I'm going to follow up on that bullet. I'm going to find out and have it summarize what happened. So I'm using my menu system that I wrote into it, I type 1.7. And I told it, if I type in an identifier of a bullet, you summarize the conversation related to that bullet. And it comes back and it says, in the email thread, the team discusses the room and schedule for the course. Charreau suggests the Tuesday/Thursday 9:30 AM time slot, but the only available room at that time is Engineering and Science Building room 048 with a capacity of 60. Jules approves of this room choice, mentioning that it's a good size for the first run of the class. The conversation does not provide any further details on the room and schedule, so it seems the team has agreed. That's wild, right? Because not only is it telling who said what, but it's also saying only two people weighed in on what room and time, and nothing else was said, so it seems that everybody's agreed. So it even is intuiting and understanding that, because nobody else said no, nobody else said yes, it's probably the right room, but it's also pointing that out. That's a really sophisticated, complex task. And how did I get there? All I did was I gave it some text, a block of text explaining my program. I've programmed in a prompt. So there's going to be two huge revolutions in computing. One is that generative AI has all these unique computational capabilities, like when it looked and it intuited and understood that, hey, nothing else in this email chain talks about or agrees to that room, so it seems that the team has agreed, that's a really complex capability. Or when we took that unstructured text where we said, my name is Jules White, I'm teaching a course on prompt engineering, and it turned it into CSV and it correctly semantically mapped it to the right columns. Those are unique new capabilities. And the second important thing is it's going to be a new interface to computing, 'cause I can go and control computing with natural language in completely new ways. I can go and generate source code for programs and run it, but I can also, just within a prompt, write an entire application and begin doing things that are useful and valuable to me, like summarizing my email. And we're doing all of this off of very simple tools right now, and the tools are going to radically explode in the coming year and months ahead.