Artificial Intelligence has become an integral part of our everyday lives, changing how we communicate, work, and even make decisions. And one particular technology has captured the attention and imagination of people worldwide—ChatGPT.
Developed by OpenAI, ChatGPT is an autonomous language model that can engage in dynamic and responsive conversations with humans, just like you and me.
In this episode of Tech Talk with Vodacom, we speak with Deon Heunis, Co-founder & Chief Technology Officer of Swipe iX, about the potential ChatGPT and autonomous AI has to build the future of the internet.
Listen to the episode below:
Transcript
Sharen
In this digital age, Artificial Intelligence has become an integral part of our everyday lives, changing the way we communicate, work, and even make decisions. From smart home devices that anticipate our needs to personalized recommendations which tailor your online experiences, AI is shaping our world in unprecedented ways.
Andre
And in the realm of AI, one particular technology has captured the attention and imagination of people worldwide—ChatGPT. Developed by OpenAI, ChatGPT is an autonomous language model that can engage in dynamic and responsive conversations with humans, just like you and me. Its ability to understand context, generate coherent responses, and adapt its behaviour has led to exciting possibilities and ignited a wave of curiosity. My name is Andre. And I'm Sharen. And this is Tech Talk with Vodacom. Today we are speaking to Deon Heunis, Co-founder & Chief Technology Officer and co-founder of Digital Agency Swipe iX, a company which specializes in software development, UX design, and digital strategy. Hello Deon
Deon
Hello Andre, thanks for having me.
Andre
Good to have you. Give us your background in autonomous ai.
Deon
In a digital agency, the way that we've always employed AI was for predictor engines. So for example, if you had to read a news article, what would be the most relevant related articles to that? Or personalization engines. So pre-ChatGPT, it was very much around personalization, predictor engines, fraud detection, monitoring and so forth.
Sharen
What are some real world applications of autonomous AI that have been successful throughout the tech industry? We all know Tesla self-driving cars.
Deon
That's a great example. Facebook, your personalized feed, when you go into Facebook, all the information you see is all driven by AI. In the banking industry, they employ AI for fraud detection. So it safeguards your account against, malicious activity and in the medical field as well. From diagnostics to personalized medicine, AI is being employed across the board in various industries.
ChatGPT has really just sort of created excitement, so to speak. Yeah, I would say AI has been around for, for many years. It's never been as popular as now and that's thanks to ChatGPT, but we've been using AI since I can remember.
Andre
So speaking of ChatGPT, how, how different was that sort of approach in, in terms of, of the use of AI in comparison to what it was used for in the past?
Deon
In the past, we would use AI for singular purposes, so whether that's fraud detection, whether that's a predictor engine or a recommender engine, it would always just do one thing really well. Where chat GTP is more of a conversational. AI model where it understands you, understands your intent, and know how to reply to you, and you can apply that across various use cases. So it's a lot more powerful and versatile in its application.
ChatGPT stands for generative pre-trained transformer. So what it, what it basically does is it understands the intent of your sentence. It knows what words are the most relevant. It would then use those words and compare it to all the data it's trained on in order to predict what it should reply with. So a good example is if I had to ask it to list all the famous museums in Paris, it would then look at that sentence. It would pull out museums and it would pull out Paris. It would take those two words and compare it to all the data it's been trained on, and then it would predict the sentence. What it needs to return. The technology was developed by, by Google in 20 20 16, I think. And that methodology has resulted in in the birth of ChatGPT.
Andre
If AI is pulling its information from external web sources, is there a danger that it would eventually start repeating itself?
Deon
I don't think so. It depends how you interface with ChatGPT. It could answer you in the style of a five-year-old, or you can ask it to pretend like it's Einstein and reply to you as if it's him. So it all depends on the question you're asking and the context it has on those questions. If you prompt it in the, in the correct way, you can get an infinite number of different responses depending on how you give context to it, how you ask it to, to, to answer you and what type of question that you ask of it.
All of the information that Chat GTP is trained on is accessed via public information on the internet. It's not a live model. It doesn't connect to the internet. There, there are some plugins that you can use now in order for it to be real time. You often you would see when it replies, like, I've only got information up to 2021, for example. And these models can live completely offline. You can, you can even install it on your, on your own computer and start asking questions. So it's, it's based on public information on the internet. Um, think about Twitter, like Twitter's public read. It is public, all the public websites in the world. So it's trained on almost the entire internet. So it's basically like a snapshot or a time capsule of everything. It was on the internet up until 2021. Well, the, the ChatGPT 5 model was up until 2021. I think ChatGPT four is obviously more recent, but yeah, I would, I don't know if it's the entire internet, but it's definitely a large, a large chunk of the internet.
Andre
So is there like a future model where it integrates with live information, live feeds, and. With information doesn't just stop at a specific year.
Deon
Yeah, absolutely. As these models grow in sophistication, it'll definitely be, it'll still be treat pre-trained on all the information. Cause it, it can't generate that information by first going to access the internet. But any up-to-date information that it would need to get from the internet.
Sharen
Is ChatGPT and other AI tools secure to use? What are the legalities around the content that it produces then?
Deon
Currently AI is completely unregulated, so there's no framework that specifies. Legality, how you can use it, why you should use it, in what way you should use it.So there's a massive movement and momentum being gathered for regulating this industry, um, in order to, to create some frameworks around its usage. So if you do use it, you need to use it in a responsible manner. You can't just ask a question and copy the, the response verbatim you would need to do your own validation. So there's still a massive amount. Of human intervention needed in using ChatGPT. So technically as a writer, my job is safer now. It is safe. Absolutely. I think the best way to use chat GTP is not to ask it. I mean, if you're talking about writing, so imagine you are a block writer, you're writing articles for, for websites.
The best way to use it is to ask ChatGPT to maybe come up with thin topics that might be interesting. Um, you would then take those topics and you would ask Chatt P to create an outline for the blog for you. From there onwards, it's still up to you to go and write the actual article, but where chat gdpp is fantastic is that it can s spawn new ideas. It can create angles to an article that you might not have think of, and it creates a quick sort of blueprint on how to go about writing this article for that specific topic. There's a fair few limitations. You might have heard about the word tokens, so tokens. Is the memory bank of ChatGPT.
So imagine you talking to the cleverest person in the world, but they've got a memory of five minutes in terms of responsibility and liability. It's still hallucinates various answers that are completely not based in, in fact, we still don't know why these AI engines reply the way that they do. You can't delve into the logs and, and analyze why it gave the answer that it gave. Um, but we still need humans. We still need to be able to write authentic content. So since ChatGPT has sort of come to the spotlight, many little other AI models have also come out that, that do other things. Midjourney is one I think of.
And then there are others that claim to be able to do, for instance, long form, right? How do they differ? I mean, if chat gpt is sort of um, all encompassing, how do they differ? How do they do what they, those those ones do what they do. ChatGPT was just the first to go public. Like these are models were in the background. But because they are still unregulated cause they are still unpredictable, um, it was probably a bit too early to release it. So chat DP kind of led, opened the, the dam walls. Um, and then since then, all these other companies had to go public in order not to fall behind. I mean, O Open ai, the, the share price completely rocketed when they launched this.
Andre
You mentioned an an AI race. You and and swipe have also joined this race because you are currently working on, on a module called Story station.ai. Tell us a bit about that.
Deon
Story Station was, was kind of born out of frustration more than anything else. The ChatGPT interface is very singular. You can only have one conversation at a time. You can only ask one question at a time, and also you, you can't save that question to reuse it for a different project, for example. So the reason we created story station was to template ize workflows that can benefit from from AI content generation. So a good example of this is think about a real estate agent. Um, they've got a bunch of properties they wanna sell, but for each property, they've gotta create a listing. They might want to create a Facebook post. They might want to create a tweet, they might want to create a Instagram, they might want to create a mailer for that listing and send it out to people. So it's, it's one workflow. And if you want to do that currently in chat, dp, you've gotta paste the information about a property, and then you've gotta ask it five questions in a time and wait for each individual output. So you've basically gotta do each and every one individually. Individually. Facebook post, the Twitter post, and exactly the mail. Uh, Yes. Um, so how does, how does Stories station AI change that? So the way that we do that in stories station is you can, you can save templates with pre-trained prompts.
So my template for a real estate agent would be, create the listing information for me, generate all the social media content around that listing, also create a mailer for me, and then by using that template, Every time I've got a new property, I'll just add the property information. I'll apply that template and it will match, produce all that content for me in one go. So it completely automates certain tasks. In addition, you can also generically use storage station as a ChatGPT alternative. There's a general chat that you can just talk to it, but you can also talk to the context of your, of your project. So if you want to ask additional information on this property, you could just talk to it and it would already have all the context.
So if you think about Gmail, if you think about your mail client, instead of typing out your mail, there's going to be an AI button which says, generate this mail for me based on. The information that I received, if you are replying to ml, it's going to be everywhere. There's no way that we are going to get away from this, and the possibilities are endless.
