ChatGPT, the advanced chatbot developed by OpenAI, has received a lot of media attention since its launch in November 2022.
Since artificial intelligence is an integral part of Uzinakod’s daily life, and ChatGPT is still very much a topic of conversation, we wanted to get our experts’ opinion on the matter. In this two-part article, our curious and passionate colleague Sophie shares her initial thoughts on this revolutionary chatbot while also discussing its limitations. As a bonus, she also provides a wealth of information on artificial intelligence in general.
Before we jump in, can you introduce yourself and tell us about your role at Uzinakod?
My name is Sophie Gonzalez, and I’ve been an AI designer at Uzinakod for over a year and a half. More specifically, I help companies adopt AI solutions. This involves building artificial intelligence tools, but also enabling companies to optimize their use. For example, we help companies automate their processes and restructure their information. When people think of data, they often associate it with a simple Excel spreadsheet. In reality, data is information. It can take many forms, including videos, images, audio, and text. It allows us to work on important projects while saving time on other tasks.
Can you briefly define artificial intelligence and explain why we’re so passionate about it today?
I’ll start with an academic answer. Artificial intelligence is a field of research that aims to build algorithms that can reproduce human logic to the best of its ability. These algorithms can therefore perform tasks without the need for humans. If you asked ChatGPT, it would probably give you a similar definition.
I think it’s interesting that it’s called artificial intelligence. It raises the question of what the definition of intelligence is, and, unfortunately, we’re not yet capable of defining it either for humans or machines. So, we need to keep that in mind. In the end, we all have a more or less vague idea of what intelligence is, which factors in a number of different aspects; it’s exactly the same when it comes to machine intelligence.
As to why it’s so intriguing… Honestly, I think people throughout history have always been both fascinated and frightened by what they don’t understand. When the first automated machines came out centuries ago, there were people who thought they should be burned at the stake!
It’s more or less the same with artificial intelligence. It’s a bit mesmerizing and exhilarating, but at the same time, there’s this fear of “is this going to steal my job.”
Can you briefly explain what generative artificial intelligence is?
Generative artificial intelligence is designed to predict possible outcomes. More specifically, it’s trained to identify patterns and connections within the data in order to generate new and coherent content. In the case of image-generating AI, the tool is given a description and will try to generate a composite of the terms provided using its training.
It has its own dataset consisting of, for example, a huge number of images associated with either very simple or more complex image description models. It then memorize everything and learns how to break down that image into multiple small pieces in order to understand its components. When you ask for a new description, it will use its newfound knowledge to create and provide a new composite. In the case of text, it’s a lot like the original language models: the first editions of ChatGPT were designed to predict the last word. You’d give it the beginning of the sentence, and it would try to guess the end. Obviously, LLMs have evolved a lot in terms of length and consistency of answers, but it’s important to remember that, by design, these models seek to generate probable text rather than enumerate facts.
Speaking of generative artificial intelligence, there’s a lot of buzz around ChatGPT right now. Can you tell us more about it? What exactly is ChatGPT and how can it be used? And why is there so much hype surrounding it?
ChatGPT is a generative AI specialized in Natural Language Processing (NLP) tasks. In other words, it specializes in text and text generation. Its different from earlier version that simply predicted the next word as it’s been trained using a very large number of parameters.
What’s a parameter? Well, it can be compared to a mathematical equation: y = ax + b. You give the AI the y and the x, and you ask it to provide the a and the b. Normally, you’d only have two parameters. However, you’ll need to input numerous parameters to carry out a more complex task, such as imitating a human. This gives you an idea of the complexity of the equation. As an example, I think that for ChatGPT3, there was almost 180 billion parameters. Again, equations would be incredibly complex.
Have you tried ChatGPT? What did you think of it? Especially compared to its more advanced version, GPT4!
ChatGPT has improved a lot. You can see how good it is. I think that’s why people are so excited about it. When it comes to language proficiency and contextual understanding, ChatGPT could easily pass for a human. That’s why it scares a lot of people whose job it is to create written material, like content writers and journalists.
On the other hand, it used to be really bad at completing tasks that required proper reasoning or computational skills. It could say things that were completely off the mark but in a brilliant and very “confident” way. It was quite entertaining. I think the new version, ChatGPT4, really improved on that. It’s gotten better at deducing simple math problems because it’s been trained to do so.
I think it’s also becoming more complete in the sense that you don’t have to interact with it with text only. You can also use ChatGPT with images. That’s the multimodal side of it, although that feature hasn’t been released for public testing yet.
Finally, we’re able to provide the AI with more context (prompt length), which is a positive change from a practical standpoint. In the past, you couldn’t ask ChatGPT for a one-page article. It just didn’t have enough words to understand the context. You would’ve had to break it up into several pieces that would eventually be forgotten.
How do you use ChatGPT? What features do you find most interesting?
The most interesting thing for me is that it acts as a sort of colleague. You can think of it as a co-worker whose labor doesn’t cost anything. For example, you can ask it for advice on how to improve the code you just wrote to make it more efficient.
I could’ve asked what its answer would’ve been when I was preparing for this interview. That could’ve given me a perspective that I might not have thought of. It’s really about consulting it to get its opinion. I would never take its answers at face value, especially since the they’re not always well sourced.
Find out more about the limitations of ChatGPT and AI in general, and how they’ll impact the workplace of the future in Part 2! 😊
