Dr. Stavros Vassos, CEO of Helvia, talks to ATHENS VOICE about the sophisticated ChatGPT platform that provides answers to almost anything, using artificial intelligence.
Below is a translated excerpt from the original interview in Greek.
ChatGPT, a natural language processing model developed by OpenAI, is designed to conduct conversations, answer questions and provide information in a natural and conversational manner, resembling human writing. It was launched in November 2022 and has attracted attention due to the detailed answers it provides.
AV: Is ChatGPT the replacement for Google?
Stavros: Let's see how Google works, and then see how a GPT system works, as you can ask questions to both of these platforms. Google has indexed all the files, texts, posts on the internet and if you ask something, it will show you the most relevant ones. If you ask, "how do I make a recipe for pancakes?", since you've used the words recipe and pancakes, it will find information that already exists, and it will provide the posts that are most relevant to what you asked for; it will not generate its own information. This is the job of a search engine anyway. ChatGPT on the other hand, and the tools of this generation, are called generative because they produce text, and they are called pretrained because they have respectively retrieved posts from the internet and "studied" them, and based on those they generate their own response. They use these posts and try to understand the hidden patterns underneath; we haven't indicated which patterns they should find, but we ask them to find as many of them and as good as they can, so when we ask a question they can answer based on the patterns found. What's the difference? For example, while Google indicates that you can find the answer in these five posts, GPT creates an answer on-the-fly based on the patterns it has stored. It's as if it has compressed all the answers – even if it wanted to, it can't go back and see the 10,000 posts about pancakes, but it has kept enough information that it can tell you: “A recipe for pancakes is this and that...". The bottom line is that the one relies on the actual posts while the other relies on what it thought was useful to store to be able to create answers to any question.
AV: How does ChatGPT benefit us? How will it change our lives?
Stavros: Artificial intelligence is not just that – ChatGPT is a huge hype because it's easy to see and play with. This is a part of AI that deals with generative stuff, e.g., creating a text piece. Does it seem absurd to write a pancake recipe if you've seen one million pancake recipes? Does it seem unreasonable if you've seen one million marketing messages on a certain topic to create a new one? ChatGPT has read all the marketing messages in the world until the end of 2021. It has read all the books available online, so - yes - it can write a book and - yes - it can write a news story. If you ask it now to write a news story about incidents that have happened in the center of Athens today, it will write it. Now, the post itself will look like many other articles it has read.
There are other systems that you can ask them to create a drawing of a table and 2 frogs playing poker and they do it, without having seen this anywhere else before. This is another widely used application of generative AI.
There's another part of AI, based on similar technology, that's about categorizing and identifying things – for example, in medicine. Let's say: "I've got these x-rays, is there any indication of some disease", or – the most ambitious example – "take this x-ray and tell me if there is any sign of cancer".
The list of use cases goes on. For example there's progress in finding faults in a factory production line. There are systems that study photos of the right products and compare them and when they see something that isn't right, they note it down. There are many things in AI that are leveraged in many ways – the generative part is just one case that is easier to appreciate in a broader scope.
And regarding the benefit: Consider an office environment and the task of putting a pile of documents into categories. If you have a junior assistant with no experience and you give them some guidelines they can easily do the job, it's kind of "non-creative" work. In these types of jobs AI is doing particularly well.
AV: Is the human factor at risk? Will jobs be lost?
Stavros: There are two answers that I really like that have been given by others. The first one is that these tools can also help beginners to learn – the learning curve with these tools is smoother compared to self-paced learning. And the second one, when one of the leaders of AI was asked about the automated generation of images by AI and whether machines will replace artists, he said that machines will replace artists who do not use such systems to assist them in their work.
The full article in Greek is available in ATHENS VOICE.