ChatGPT (Generated Pre-trained Transformer) became a hot topic around the world as soon as it emerged. More and more people are curious about ChatGPT, and more and more people are using it. ST once investigated what it is and what makes the world so enthusiastic about it, what influence it has, and how it differs from existing AI (Artificial Intelligence). Furthermore, both the positive and negative aspects of ChatGPT were addressed in the article. People have also explored how to use ChatGPT and how to use ChatGPT in the future, dealing in depth with ChatGPT. If you are curious about ChatGPT, let’s find out with ST through the article..............Ed
The emergence of a 'super-large AI (Artificial Intelligence)‘ called ChatGPT (Generated Pre-trained Transformer) is shocking the world. ChatGPT is a generative AI that has entered the realm of ‚creation‘, which was considered a unique domain of humans. It shows a scary and amazing ability by creating something that do not exist in the world. Furthermore, it outperforms existing Al models in other tasks, such as understanding the context of language, correcting errors on its own, and generating convincingly human text. It is necessary to consider the positive and negative effects of ChatGPT, and consider how much technology will be applied to our daily lives. With the ChatGPT craze, the world has already begun a war of AI technology hegemony, and ST intends to analyze the various effects of ChatGPT, and discuss the implications of such a technology.
- What is ChatGPT?
The conversation-specialized artificial intelligence paper released by Open AI on November 30, 2022, is a chatbot, a large-scale artificial intelligence model made by Open AI. Chat is short for Chatting, and GPT represents the first letters of ‚Generated Pre-trained Transformer‘. This is a service that allows users to enter text into the chat window and converse. It has become a sensation with more than 1 million users a day in just five days of publication, because it is clearly different from existing AI as it can answer questions, translate, compose songs, and code. ChatGPT uses pre-trained mass-generated converters in a vast database of millions of web pages to generate human-like conversations. Moreover, human feedback- based reinforcement learning techniques have been applied to derive answers that meet users‘ intentions and needs. In existing AI learning data, human work is small, or does not exist. However, ChatGPT increases the quality of final output based on their feedback by placing a ‚human‘ learning guide called 'Labler‘ in the middle of AI learning data.
- ChatGPT’s advantages
1. High accuracy of natural language processing
It is possible to provide answers to users‘ questions with high accuracy because it is learned based on the vast amount of web data that exists on the web.
2. Interactive interface
It can interact with the user in the form of a conversation, which can improve the user experience. It also analyzes previous conversations with users, and provides information in context.
3. High processing speed
It can provide a quick response to a user‘s request at a high processing speed. Therefore, rapid prediction and generation work are possible.
4. High accuracy and diversity
It can support various tasks, such as natural language generation, Q&A (question & answer), information extraction, and emotional analysis. Also, it is trained based on a large amount of data, so its ability to cope in general situations is also good.
- ChatGPT practical use cases
1. The field of education
Education companies can use ChatGPT to provide immediate academic advice and answer students‘ questions. Moreover, one-on-one interview coaching, self-introduction writing, and poetry writing are possible. This can help educational companies improve student participation and access to academic advice.
2. E-commerce and financial services
Using ChatGPT, e-commerce companies can provide custom product recommendations, answers to customer inquiries, and customer support. This could help e-commerce companies increase sales, reduce shopping cart neglect, and improve customer retention. Financial service companies also use it to provide customized financial advice to increase customer satisfaction and improve customer experience.
3. Medical care
Healthcare companies can leverage ChatGPT to provide immediate medical advice, answers to questions about their health status, and support scheduling appointments. This improves patient satisfaction, reduces latency, and improves access to medical advice.
4. Travel and Accommodation
Travel and accommodation companies can use ChatGPT to provide customized travel destinations and hotels or multiple attraction recommendations, and answer customer inquiries. Since the adoption of ChatGPT, travel and lodging companies have increased their reservations, reduced cancellation rates, and helped improve customer satisfaction.
An airline provider has also adopted a group seat recommendation service through ChatGPT. As of now, the service is not without its faults; the overall booking status of the aircraft is not easily visible on one screen, and for a party to sit together, they had to individually select seats. However, it does inform the consumer of the probability of sitting apart, and tries to find seats that the party can sit together on. Customers usually tend to prefer front seats close to the exit, but consumers who value flying with their party more have shown willingness to consume back seats through seat recommendation services. This shows that conversations with ChatGPT affect consumer preferences and value judgments.
The entire world, it seems, is enamored with the power of ChatGPT, and how such technology can affect business and leisure. However, ChatGPT is not perfect, nor is it a pure force of good. ChatGPT, like anything else, has shortcomings and downsides.
- Chat GPT’s disadvantages
1. Unable to learn in real-time
ChatGPT is unable to collect and learn text data in real-time. Currently, it has only learned data until 2021. For ChatGPT, the president of the Republic of Korea is Moon Jae-in, and the Russia-Ukraine war has yet to break out. If you ask ChatGPT who the president of the Republic of Korea is, you can get the answer, “I don’t know the current president.” Knowledge evolves; it is not fixed. This means that ChatGPT cannot replace search engines such as Google.
2. The limit of memory
ChatGPT has memory limitations, making it impossible to have consistent conversations with users. According to an official answer from Open AI, ChatGPT can remember approximately 3,000 words in conversations with users. Exceeding this limit, ChatGPT forgets previous questions and answers. In other words, to revive previous memories, you must enter past questions and answers together with new questions every time. The limitation of ChatGPT’s memory is an essential problem.
3. The problem of copyright
ChatGPT is widely used is in coding. It instantly creates code that contains the programming language and the intention of the user. Because it also finds the error in the code easily, people are forced to visit ChatGPT frequently. This is where the copyright problem of ChatGPT arises. Copyright problems had risen before with services like copilot or platforms like Github. With ChatGPT using code from such sites without permission, the problem of copyright can only worsen.
4. The possibility of the monopolization of knowledge
Knowledge is neither static nor monolithic. What you learn can never be the whole or complete truth. This means that being educated on the process of learning is even more important than simple knowledge. However, ChatGPT simply summarizes its data and gives you an answer, without teaching how you can learn by yourself. Therefore, relying on ChatGPT for learning can leave gaps in your knowledge. Even if the learning process is simplified and made more efficient, you cannot be exposed to different opinions and ideas, thus restricting your ability to think critically.
What are the positions of relevant personalities on the matter of ChatGPT? Apple co-founder Steve Wozniak and Tesla CEO Elon Musk claimed that developing AI systems that outperform GPT-4 was dangerous, and proposed stopping AI development for six months, citing potential dangers of overdeveloped AI. On the other hand, other figures refute this claim. Bill Gates claimed that suspending AI development does not solve the problem, and predicted that there is a big advantage in ChatGPT’s technology. In addition, Yann LeCun, the chief AI scientist at Meta, argued that stopping AI development is ridiculous, and ideas of AI being dangerous is fear-mongering.
Then how should we use ChatGPT in the future? According to the New York Times, professors are troubled with ChatGPT helping American college students write reports. In response, New York City banned school Wi-Fi networks to prevent middle and high school students from using ChatGPT. However, the technology is not going anywhere. Open AI is planning to release its follow-up to GPT-4, and Google, as well as other companies that built Lambda, will also start releasing their own generated AI. Since it is not possible to stop progress in AI, such a ban will become meaningless as the tech becomes more and more integrated with our lives. In such a world, the ability to not only know the answer, but to ask the right question becomes paramount. This is because ChatGPT can offer what is called a hallucinating answer. For example, if you ask for the ingredient for a tomato pasta drink, the best answer ChatGPT can offer is, ”I don‘t know the answer”. It is because such a thing does not exist. ChatGPT is not a magic 8-ball; it cannot offer a wise answer to a foolish question. In fact, a study conducted by Professor Terrence Sejnowski of UCAL on behalf of the Salk Institute for Biological Studies in the U.S. showed that the pattern of ChatGPT‘s answer varies, depending on the facts the question presupposes. According to Professor Sejnowski, ChatGPT reflects the level of knowledge or intention of the one posing the question in the process of constructing answers. “It’s like ‘The Mirror of Erised’ in the famous fantasy novel, Harry Potter,” he said. This mirror, which appears in the novel, reflects the deepest desires of those who explore it, never yielding knowledge or truth, only reflecting what it believes the onlooker wants to see. Moreover, in this study, it was found that ChatGPT reflected the various emotional states of the person asking the question. If chatbots receive ideas that are emotional or philosophical, they will respond with answers that are emotional or philosophical. Professor Sejnowski said, “They can be wonderful tools, but only if you know how to use them. Otherwise, you end up being misled and be caught in potentially emotionally disturbing conversations.”
‘Knowledge’ is essential to analyze, infer, and evaluate information. Knowledge is also essential for critical thinking; it should not be overlooked that we must have a standard by which to judge reality to be critical. ChatGPT is a tool; how it is used, and how useful it can be will depend on the user.
Yang Su-rin (ST Reporter)
Lee You-bin (ST Reporter)