ChatGPT (And Other AI Language Models): What’s All The Hype About?

Author: News Bureau
Posted: Monday, April 24, 2023 12:00 AM
Category: Pressroom

Macon, GA

If you haven’t yet paid much attention to ChatGPT, or AI language models in general, you will soon. People are increasingly using AI language models in a wide range of applications, such as chatbots, virtual assistants, and content creation. AI language models have the potential to significantly transform the way we interact with technology and each other – but with their use comes plenty of controversy and possible drawbacks. We asked #MGA’s Dr. Myungjae Kwak, IT professor, to discuss the potential advantages and pitfalls of AI language models.

What is an AI language model, and how does it differ from other types of AI models?

An AI language model is a computer program that can understand and produce human language. It is like a smart assistant that can read text, listen to speech, and generate responses based on what it has learned from vast amounts of data. They learn from enormous amounts of data, like books, articles, and online content, and use complex algorithms to recognize patterns and make predictions about what words should come next in a sentence. By doing this, they can simulate human-like conversation and assist in various tasks such as language translation, content generation, and answering questions. There are many different types of AI models. An AI language model is different from other types of AI models because it focuses specifically on understanding and generating human language. Other AI models are typically designed to perform specific tasks like recognizing images or predicting outcomes, while language models mainly work with text and speech.

What are some of the popular brands of AI language models? We’ve been hearing a lot about ChatGPT, but there are others, correct?

Since OpenAI launched ChatGPT in November 2022, many companies have been racing to bring out their own AI language models. Of the many AI language models currently available for general use, ChatGPT (Generative Pre-trained Transformer) based on GPT-3 and GPT-4 developed by OpenAI, Bard based on LaMDA (Language Models for Dialogue Applications) created by Google, and LLaMA (Large Language Model Meta AI) created by Meta (formerly Facebook) are the most popular brands. Especially, 1 million users visited ChatGPT in 5 days of launch, and by January 2023, 100 million users had tried it, setting the record for the fastest-growing digital platform. In February 2023, it had reached one billion visits, and by March, 1.6 billion visits had been made. This huge boom has influenced numerous other organizations to develop and train their own AI language models, as well as inspiring software companies and engineers to explore the potential of using AI language models to create practical applications.

What are some of the key applications for AI language models, and how are they being used in industry and research?

Given the impressive performance of AI language models trained on large datasets, their widespread adoption in industry and research has already begun. A number of real-world applications based on AI language models are now being used for customer service, marketing, e-commerce, healthcare, software development, journalism, office work, and many other purposes. For instance, chatbots and virtual assistants that can understand and generate natural language are being extensively used to lower customer service costs. Microsoft recently released Microsoft 365 Copilot, which combines OpenAI's GPT-4 with Microsoft 365 Office Apps such as Word, PowerPoint, Excel, Outlook, and Teams to assist users with emails, documents, meetings, and data analysis. Within Word, it can generate a draft document for a particular topic or modify the document according to a prompt. For Excel, it can analyze a given data set and present the analysis results in the form of text and graphs. It can also create PowerPoint slide decks based on given documents and data. For Outlook, it can not only quickly answer routine emails but also retrieve emails related to a particular one and draft appropriate responses. Other real-world applications such as machine translation between languages, sentiment analysis for market research or social media monitoring, content generation for marketing, spam filtering, personalized language learning and tutoring tools, and tools for the disabled like text-to-speech and speech-to-text are currently available.

What are some of the potential drawbacks to the use of AI language models?

While AI language models provide many potential benefits, there are also potential problems to the use of AI language models. First, if AI language models are trained on socially and politically biased or incorrect data, they can generate biased outputs. Also, they may generate content that is factually incorrect, misleading, or outdated. The models are generally trained on a diverse range of datasets, but we should know that they may not always produce the most accurate or up-to-date information. Second, even though they seem to understand human language, the models inherently learn by identifying patterns and statistical associations in the training data. Therefore, they sometimes provide nonsensical or irrelevant answers without a deep understanding of the content and generate content that is offensive, harmful, or inappropriate. Third, AI language models collect and learn from large amounts of personal and organizational data, which could be problematic when it comes to privacy and data protection. Finally, other unknown problems and issues could be raised in the future. It is very important to carefully consider their limitations and potential drawbacks when using them.

What are some of the future directions and trends in AI language modeling, and where do you see the field heading in the coming years?

Researchers and practitioners in the field are actively working from many different perspectives. First, many of them are exploring to create more powerful AI language models by increasing the size and complexity of language models by using larger training datasets, more sophisticated algorithms, and new hardware and software architectures. Second, they also are studying to capture the true meaning of text in context to improve the accuracy of language models even more. Third, with the growing demand for multilingual natural language processing applications, researchers are also working on language models that can more effectively process and understand multiple languages. Fourth, generating digital content such as scripts, images, and videos can be more efficient and effective with the help of AI language models. Many researchers and practitioners are actively developing AI algorithms that can generate high-quality text and digital content. Finally, some AI experts, tech entrepreneurs, scientists have raised concerns about the rapid development of AI language models, and they recently called for a pause on the development and testing of AI models more powerful than OpenAI’s GPT-4 until their risks are properly studied.  

To sum up, AI language models are powerful tools that can automate tasks and provide assistance in a wide range of fields. It is anticipated that they will be used widely in our daily lives, changing the way we live. It is expected that they will replace some jobs, but also create new ones. However, they can also be biased, generate fake content, and be misused for malicious purposes. It is essential to use AI models responsibly, with human supervision and ethical principles in mind, to mitigate potential risks and ensure positive results.

Dr. Myungjae Kwak is a professor of information technology at MGA's School of Computing and holds an MS in computer science and a Ph.D in information systems and technology. Currently, he is teaching a range of courses in software engineering and data analytics.