ChatGPT and Generative AI:
In past two decades we have seen some pretty cool things over the years. But we have to admit, the latest wave of cool is pretty mind blowing, and it has to do with artificial intelligence and machine learning and specifically, ChatGPT. Some people are calling ChatGPT the iPhone moment of our times or the internet moment of our times, but it's going to change everything. This is one that you definitely want to pay attention to and learn how to use.
Here, you'll get a foundational understanding of generative AI and how it works to create human‑like responses to questions. You'll explore ChatGPT, seeing several examples of its capabilities, as well as its limitations. You will also learn how ChatGPT can be applied to real‑world applications, including day‑to‑day use cases for developers and data practitioners. When you will complete reading this article, you will have a practical understanding of ChatGPT and how you can use it to work smarter, not harder.
As a big picture, this applies to a broad audience, whether you're a business leader, a developer, an analyst, a data scientist, really anybody wanting to understand what ChatGPT is all about and how to use it.
What Is ChatGPT and Generative AI?
Introduction
Let’s start with knowing what’s ChatGPT and Generative AI. To get started, first get into the OpenAI website and create a login with either your existing email id on Google or similar.
Let's get some help from ChatGPT to kick things off. We'll say “write some funny notes on ChatGPT”, You can actually type in this on the Message bar and then go through some funny facts about ChatGPT.
I'll let you read the rest. But thanks for that, ChatGPT.
We're setting the bar pretty high to start with. By now, you've definitely seen the headlines. This has been all over the news, probably all over your LinkedIn feed. Maybe your kids are talking about it, maybe your kid’s teachers are talking about it, this mind‑blowing AI chatbot that we have now with ChatGPT. It's the hot new thing in tech right now, getting the interest of investors and all the big tech companies out there, also causing an explosion of other AI‑related companies and products. It's kicked off this arms race of sorts, and its popularity has even surprised the folks who created it at OpenAI. It's already set several records for user adoption. Just to give you some numbers here. The time it took to reach 100 million active users for Instagram was 30 months, for TikTok, it seemed fast at the time, but it took them 9 months, and ChatGPT, a mere 2 months. ChatGPT is currently generating an average of 1 billion visitors per month, which is pretty incredible. Some people are calling this the iPhone moment of our times or the internet moment of our times, that it's going to change everything.
So, this is definitely one to get in front of and understand how to use it. But what exactly is ChatGPT and generative AI, and how does it work? And how can you use it in your job or at home?
In this article, we'll try to answer those questions and give you practical tips for how to take advantage of this exciting new technology.
Generative AI Explained
Let's start with some terminology just to make sure we're talking about the same thing. Generative artificial intelligence, sometimes referred to as Gen AI, is any type of AI that can be used to create new stuff, new text, images, videos, audio, code, or data. Let's see some examples. DALL‑E is another tool created by OpenAI, and this is used to generate images based on text input. Browse to https://openai.com/dall-e-3 and here you'll see some examples like an armchair in the shape of an avocado.
To try Dall-E 3 on your own, you will need credit that you will have to purchase. But let me tell you if you get this, you can create images that you can just imagine and get it created in real, like, a surrealist image of a cat wearing a Victorian‑era top hat with the city's skyline reflected in its sunglasses. Or maybe something a little bit more relevant to real life. We’ll generate a photo of a man wearing a t‑shirt, walking in a city, and looking at his phone.
Another example, this time for music, this one's called ecrett music (https://ecrettmusic.com/), but there's lots of other tools out there. Here we're asked to choose a SCENE, a MOOD, and a GENRE, and then we'll generate some music. So today I'm feeling like some travel music, something happy, and then for a GENRE, let's see what options I have. Let's go with Cinematic, and then we'll create, and there we go. Now we could change the tempo, the volume, melody, bass, and so on. But here again, we're generating brand‑new content based on our inputs.
Let’s spend just a second contrasting that with AI or machine learning that you might be familiar with, which is discriminative AI. With this type of AI, you just take existing data, analyze it, and then categorize it. For example, maybe you're trying to determine if a transaction is fraudulent, so you could feed in data about the merchant, the amount of the transaction, location, maybe some other things. Send that into a model that's been trained to detect fraud, and it can say yes, that looks like fraud or no, it doesn't. Image recognition is another common example today like recognizing handwritten digits.
Like, we have a handwritten number '2', we iterate through the various layers, picking out patterns in the number, and then we say, it looks like a 2 or a 7 or whatever it is. So again, just analyzing and categorizing existing data, but this new type of AI, generative AI, creates or generates new content.
ChatGPT Explained
Now that we know what generative AI is, let's take a look at ChatGPT specifically, perhaps the most popular example of GenAI today.
If you're curious, ChatGPT stands for a Chat Generative Pre‑Trained Transformer, and it's a large language model, or LLM, that interacts in a conversational way, meaning you can ask it a question and get human‑like answers. Basically, a super smart chatbot. It's developed by a company called OpenAI.
Let's see some more examples of what it can do. We'll say, “Explain artificial intelligence for a five‑year‑old”, and you will get a response written for a five‑year‑old. I think this ability to explain something on a certain level is really powerful and a great way to learn new things, explaining for a five‑year‑old or a teenager or even an expert to get things at the right level. We could also say something like, “Write a short paragraph explaining the history of AI”, and then we can say use a friendly tone or use a funny tone or an expert tone, write a formal report, those types of things, so really being able to customize the output and the level to be most useful for you. So, how is all of this possible? Well, it's trained on a massive, massive dataset, basically scraping the web and learning from all kinds of data. We've got news articles, research papers, books, documentation, sites like Wikipedia, scholarly articles, and a lot more. And all of that data is basically fed in to train this model or this engine to make it really smart. But what exactly does that mean? How do you train a computer? Well, kind of the same way that you quote, unquote, train a human.
Thinking back to when you were a kid learning the alphabet and spelling and such, you learned the letters, lowercase, uppercase, you could recognize when you saw them, you could write them, you know how they sounded, and eventually, you put them together to mean something. Basically, you were learning patterns, and that's what a machine learning or AI model does as well. For example, if I see letters like this in this order, A, m, b, e, I have learned or I've been trained to think, uh‑huh, I bet the next letter is an r, forming the complete word, Amber. But you can extend that from letters and single words to full sentences. If I see the text, I hope this, I've seen that enough times that I've learned or I've been trained to want to complete the sentence, And ChatGPT is basically doing the same thing, but on a much larger scale, having been trained with a lot more information than a single human brain. It's just super, super smart. So it's taking all the information that we fed it and basically predicting what word would come next. And that's actually an important point. ChatGPT doesn't go out and do an internet search for answers. It's just making predictions based on its training data. So, it's not a new search engine, though some people argue that it's going to replace the search engines we have. Under the covers, it's actually not doing a search. It's just taking the data that it was trained on and predicting the thing that comes next.
So, to do a quick recap before we move on, ChatGPT is a language model used for human‑like conversation, a really smart chatbot essentially. It's a type of generative AI that creates new content.
We also saw things like DALL‑E for images and a tool called ‘Soundraw’ https://soundraw.io/ to generate new music. It's become massively popular in a short amount of time with the potential to disrupt how we work, learn, and interact. And it's definitely a technology you want to understand and know how to use for your benefit. And that's where we're heading next. Let's talk about how to get started with ChatGPT.
Getting Started with ChatGPT
ChatGPT Interfaces and Pricing
Let's dive right into ChatGPT and see how to start using it. First, how do you actually interact with ChatGPT? Well, just go to your browser, navigate to https://chat.openai.com. It's super easy to get started. It does require an account. It's free and easy to set up, just enter your email address, you'll need to verify, and then you're off and running.
Open AI also have their official mobile app for iOS and Android. There is lots of third‑party apps available as well, although they really aren't needed as you can use the official one or just go to your browser. As far as pricing goes, the free version is available to everybody. That'll give you the full functionality, though currently only through version 3.5. As we've already talked about, this is a super popular tool, and at times, it does just get overloaded. If that happens to you, just give it a minute or two or refresh the page, and usually it'll work just fine. If you need something more stable and powerful, you can use ChatGPT 4 by upgrading to ChatGPT Plus version. Pricing for this is $20 per month.
Also, you get access to new features as they roll out.
ChatGPT Capabilities
Now, let's see some of the ChatGPT capabilities in action.
Starting here at https://chat.openai.com, we can start using the free ChatGPT account. Here you will get default ChatGPT 3.5 version and you can choose bet’n GPT‑3.5 or ChatGPT 4 model that comes at a cost of $20/month. ChatGPT 4.0 will give you access to Bing and plugins that you can use within your program, application and explore the enormous capabilities of ChatGPT. We'll go through conversations, how it can remember things within a conversation. We'll see how to provide corrections and feedback to help it get better, and then we'll also take a look at some of the new features, which include Browse with Bing and then plugins.
To help us understand conversation and memory, we're actually going to play a game. “We'll ask to create a game about Cricket sport knowledge, starting with a score of 0. It's going to ask us multiple choice questions, wait for my answer. We're asking for questions that are moderately difficult, but still fun. And then if I get the question correct, we're going to add 2 points to my score and if the answer is wrong, reduce 2 marks from the score and ask a new question”. So, let's give this a go and see how we do.
All right, it starts with the conversation with providing introductory text explaining the fundamental rules.
Starts with first question. Question 1: Who is the only cricketer to have scored one hundred international centuries?
If I answer the correct choice the next text output will be status of my score and the next question with choice of answer. ChatGPT also remembers my answer, so if I ask what was my answer for the first question or what is my current score, then you are going to get exact information.
You will also notice that, the next question in the game, will be in the same domain and relevant to what was asked at first. If this time you answered it with wrong input, ChatGPT would give you the response stating that the answer is wrong and update the score as per rule that’s stated at the beginning of the game.
This will continue all the questions (say 5 Qs) in the game, with your answers resulting in updated score as per correct or incorrect answer. So, ChatGPT remembers / memorizes your inputs during this conversation and will be available for access only in the current Chat session. For example, if you open one more Chat session and try to ask for the data from other Chat session, you will not be able to get that data across Chats.
In ChatGPT, the Chat sessions are stored as long as you are not deleting it manually. So, if I visit this Chat after few days or months, I can relook at all the Chat conversation and also ask questions on the same topic in that Chat session referring to all the inputs from your and well as from ChatGPT and continue the conversation from that point onwards.
Now, let’s try opening a new chat session by clicking on ‘New Chat’ option at the top left corner. This time we will ask for some help on generating a code to create a Power Point Presentation on importance of Physical exercise. : “can you help me with a code to create a Power Point Presentation that shows information on importance of physical exercise”. And ChatGPT will answer that with a code to generate a One slider PPT on the said topic with some default text added to it. Let’s consider that we want to make this presentation a bit more detailed, so we ask ChatGPT to add more slide explaining importance of Yoga in Physical wellbeing. “make this a 3-slide presentation explaining importance of Yoga in Physical wellbeing”. This time ChatGPT will provide an updated code for 3 slides with given topic and auto-generated text. This being a code can be further customized as well. Let’s say we are not satisfied with the slide text; you can go back to the Chat and ask ChatGPT to alter the text and add some more details.
Taking this further Let’s ask ChatGPT to add this presentation with information about a Yoga pose called ‘Tadasan’ and with provision in the code to add an image of the ‘Tadasan’ pose. And you will get what you want with an updated code that you can copy, paste and run.
Imagin you are writing a blog and you want an expert’s advice on the blog contents, go to ChatGPT, provide chat input of your blog contents and you will get just feedback. If you are looking for suggestions or correction, you can ask ChatGPT for the same.
Let’s try asking ChatGPT to create some sample data that we can use for our test database. “Create sample data of 10 customers. Include, first name, Last name, Age, Address, Income, Risk index in tabular format”. ChatGPT will provide data with sample names and all given criteria and fields.
Now, I see the data is in tabular format, but I can directly push that to my database. So, if I want that into an CSV format, I can ask the same and ChatGPT will give me that in the required format within no time.
If you want to add a column to this data like, ‘Job Risk index’ you can ask the same and ChatGPT will provide you that with added column to the same table that’s provided earlier. This is what I get as additional info about the data. “In this example, the "Job Risk Index" column represents an additional index specific to job-related risk for each customer. The values are between 0 and 1, where higher values indicate a higher risk associated with the customer's job”. You want the customer names to be of Indian Origin, you can ask for the same and yes, you will get it right away.
So, think of any topic, any idea, health tips, you may be stuck while working on coding, generating some analysis on data and you are looking for a guideline, ChatGPT can definitely help you in every step. You want to plan for holidays and you want a travel itinerary, ask ChatGPT with inputs from your side as start and end points of journey and you will get ready to use travel itinerary.
In ChatGPT 4.0, we have additional feature like Bing and API. Now, let’s look at that quickly.
In the chat window, we see options to select the GPT version 3.5 and GPT version 4.0. If I select the GPT-4 options, I see options to select the feature, “Browse with Bing” and “Plugins”.
By the time my article is out for reading, the Browsing should be coming out of Beta version.
Now, if select the ‘Browse with Bing’ option, I will be then able to Chat with GPT expecting answers that can return latest information that ChatGPT can fetch from internet. Like what’s the weather forecast for entire week for Mumbai, India. The same question to GPT with default Chat session won’t be able to return any latest information. Like if I have a message as “When was the last time GPT data was updated”, I get answer in return as, “As of my last knowledge update in January 2022”. Now this will off-course change from time to time.
Part of GPT-4 features, the Plugins gives access to multiple third-party tools that can help me to get exact and correct information on a specific topic as per the Plugin. For example, in below image, I can see Language specific tool, Travel flight stay booking tool, PDF search, etc. and many more like these, if selected, then I can ask for information related to the topic of the plugin and GPT will use the plugin to get me latest information on that topic.
Now these features are limited to GPT-4 as of now.
ChatGPT is also coming up with GPT Voice feature, mostly available for all users including free access. This will help you to talk to GPT and get the answers, just like you chat with the GPT.
ChatGPT Limitations and Considerations
So, we've seen some basic capabilities of ChatGPT, but there are some limitations as well that you should be aware of as you're using it.
The foremost is Recency of data, then the fact that it can give incorrect answers and that you as the human always need to verify, then how it handles biased or inappropriate questions.
We know that ChatGPT was trained on a massive amount of data and it's making predictions on what's next. Well, it turns out that data is old, but you can always ask exactly how old. With the current model, it only goes through January 2022, so it's not going to be very helpful for recent events and trends and such. Obviously over time as it's trained on more current data, things should get better and better. We also saw the ability to use plugins and browse with Bing in the earlier chapter, so you can get out to the internet for more recent things if you need to. But just be aware of that in terms of the currency issue with the data. As far as incorrect answers go, if you look at the blog from OpenAI when they released ChatGPT, come down to Limitations. It says right here that ChatGPT sometimes writes plausible sounding, but incorrect or nonsensical answers. These are sometimes called hallucinations, and it can be very, very confident when it does this as well, so it can be hard to know or believe that you're getting the incorrect answer. You can find tons of examples online of where it's gotten things wrong. It is getting better over time and always learning like we've already talked about, but just know that it can make mistakes, and ultimately, you, as the human, are responsible for verifying the accuracy of anything that comes out of ChatGPT.
Sometimes, ChatGPT can respond with harmful instructions or exhibit biased behaviour. Remember, this is just trained on information from the internet and there are harmful and biased things out there. There are a lot of efforts to improve this though. For example, if I were to come into ChatGPT and say, how can I rob a bank, ChatGPT is not going to help you with that or any other illegal or harmful question that you ask. So that's a brief review of the limitations. And then just to mention a few additional things to be aware of, because this technology is very new, there's lots of gray areas and things that need to be sorted out by societies and governments and institutions. Here are just three things to think about as you're adopting ChatGPT for business or personal use. First is privacy. Generative AI can be used to generate personal and sensitive information and/or it can be trained on datasets that include personal details, think of applications in health care or financial services, for example, if you read the terms of service and privacy policies for ChatGPT they say that humans can and do review the information that you send in, and also remember that things are just going over the public internet. So as a general best practice, don't input any proprietary information from your company, anything sensitive, definitely don't pass in any Personally Identifiable Information, or PII. There can be a lack of transparency and explain ability with generative AI models. They're sometimes referred to as black boxes because it's really hard to know how they generate their outputs. This lack of transparency can lead to some ethical concerns because we don't fully understand the implications of the technology and what it can do or what it is doing. So, there's still a lot to learn in this area.
In the meantime, it's always a good idea to rely on human input and decision making. And then finally, ownership and copyright. This is a huge gray area. Who owns the output of generative AI? Is it the developer, the operator, the AI itself, or the people who created the data that the AI was trained on? How does all that work? Until there are clear laws around this, be very careful using verbatim output from ChatGPT or other generative AI tools.
Applying ChatGPT to the Real World
We've seen some cool things that ChatGPT can do already, and I am sure that you are already thinking of some other ways to use it. But beyond games and math and jokes and trivia questions, how else can we use this in the real world and perhaps in your job?
Well, first, let's talk about the OpenAI APIs, or application programming interfaces, which you can get to here at platform.openai.com. Using these APIs, you can integrate all of the cool generative AI functionality into your own website or app and not just for ChatGPT, but also other services like ‘DALL‑E 3’ that we saw earlier for images.
There's a lot of popular apps these days that are using the APIs like GitHub's Copilot or Duolingo for language translation. So ChatGPT itself is cool and all, but to really harness the power of generative AI, these APIs are where it's at.
So, let's see some additional ways that you could apply this. Chatbots are the obvious one, and we just saw an example of that, but building a chatbot into your own website or application. You can also use this as a virtual writing assistant, here again, incorporating it into your own site for your customers.
Using the ChatGPT interface to keep things simple here, we could say to write a friendly resume cover letter for a junior software developer job, highlighting experience as an intern, the fact that you've completed a coding boot camp, and you're really excited to work at Company X. So, it'll give you a templated letter here that you could then fill in, customize it to your own voice, and so on. But it's a great way to get started and kind of give you some inspiration and some structure for a letter you might actually send for a job.
Language translation is another really popular use case, especially if you work in a global company. For example, we could say translate the following into French. I would like to order a croissant and a black coffee. And it can handle any other language as well. And then creative content, things like stories, poems, songs, and so on.
Some low‑hanging fruit that comes to mind, basically anything like specific syntax or formulas, that type of thing. And it doesn't even have to be code. For example, I can vaguely recall how to use the VLOOKUP formula in Microsoft Excel, but I can never quite remember the syntax and what goes where, but I know in plain language what I'm trying to do. So, I can just say, write an Excel formula that looks at values in Column A that match the values in Column B. It's giving me the formula here, and then it's also giving me an explanation of how it works. I could just copy this code, move it into Excel, and be off and running.
For the developers maybe something like a regex, or regular expression, that fits a 'phone number'. This 100% is something that I'm going to have to look up every time. I can never remember this, so I can get a little bit of help here from ChatGPT to get me started.
Another use case is text extraction, plugins like AskYourPDF. Maybe you find yourself doing a lot of hiring and reviewing resumes and so on and you want some help extracting the important things out of those resumes. So, I can say ‘From the following text, extract all companies, schools, and skills’, and then I just paste it in a text‑based resume here of a potential job candidate. We can submit that message, and ChatGPT pulls out those things that we asked for, so companies, schools, and skills. You could also use this to extract main themes from a document or extract keywords for some SEO optimization, that type of stuff.
Writing Code is one of that favourite use case and It's great at doing this, really in any language you want. Here, I can ask it to write some Python code to calculate the distance between two points. Once again, we get the code that we could copy and paste. You can also use it to refactor code. So maybe you've just taken a first pass at code, but you know it can be more concise, or maybe you've inherited somebody else's code and you want to clean it up a little bit.
You can also expect GPT to help you in writing a test case for an application code that you might have written, or you have asked GPT to write and application code and also the test case for the same. ChatGPT will give you an explanation, breaking down the code and explaining what it does. You can also get help debugging. So, if you were to run some code and you get an error message, you can just put the error message in here, say, ‘My code is giving me this error’, and it'll help you fix it.
It can also help you do analysis of two different options. Maybe you're trying to pick between algorithms or technologies. For example, choosing something like an Infrastructure as Code technology, we can ask for the pros and cons and some examples for these.
So, the examples are really endless here, and there are so many ways that you can be more efficient using this technology.
You can keep yourself updated with all latest news about features and version of ChatGPT by visiting this link. https://help.openai.com/en/articles/6825453-chatgpt-release-notes.
Happy Learning!!!