The chatbot developed by OpenAI has been taking the world by storm. In the last few months it's been featured in news outlets globally and has been heralded by some scientists as being the cornerstone of AGI. So what is it and how does it work?
ChatGPT is a chatbot AI based on GPT-3 (generative pre-trained transformer), the third model of their natural language processing project. Basically, it's a technology that sifts through a massive pool of internet data and sources to reference its knowledge. Now there is a lot of different technologies and systems that go into making ChatGPT. For example, it is transformer based and it's also using a neural network. Using those technologies, it's been programmed with an unending amount of text data to understand context, relevancy and how to generate human like responses.
Ok, great, but what exactly is a neural network? What does transformer based mean? I'll explain!
Neural Network: Machine learning algorithm that is designed to work like a brain. Just as a brain has neural pathways where information is stored and functions are carried out, neural networking algorithms mimic this by learning patterns and collecting data to problem solve via neurons.
Deep Learning: It's a neural network with three or more layers. Deep learning, a subset of machine learning, attempts to simulate the human brains ability to learn, by allowing the AI to "learn" from huge amounts of data.
NLP - Natural Language Processing: It's what ChatGPT is. It's a sub-section of AI that focuses on the interaction between computer and human using language. By using algorithms and models, NLP can analyze, understand and then use language with human diction
Transformer: No, not an autobot or decepticon. This is a structure within a neural network meant for NLP processes that use tools to analyze input and generate output.
GPT: Is a transformer based language that's been developed by OpenAI. This is the language processor and generator part of the program and what generates text in a human like way.
Well, it's pretty simple, at least at face value. You put in a prompt, and you get a response. You could do something as simple as "Who was the 15th president of the United States?" to "Write a python script that will uninstall all non-essential software from Microsoft." Now that doesn't mean it's going to spit out a 100% accurate line of code. There are likely more variables there it needs to know, that you can define. For example, what is considered a non-essential software? The president question though, would very likely be correct. See, the more and more ChatGPT is used, and the more data it is fed, the more it learns. The more it learns, the more it can correct itself, adjust and reevaluate.
ChatGPT can generate text, or complete it. It can answer questions, summarize, translate, and act as a conversational partner.
Because of this, you may have seen in recent news the issue schools and teachers specifically have had with it. ChatGPT has been banned in many schools and institutions because of how many students were using it as a tool to produce essays and other written works. This article by The Atlantic called "The College Essay is Dead" covers this issue better than I ever could.