The public release of ChatGPT was the first time many people knowingly interacted with artificial intelligence (AI). Released on November 30th 2022 by OpenAI, ChatGPT is a chatbot designed to give human-like responses to prompts, but this has led to confusion about what ChatGPT is. According to Pew Research 14% of US adults have used ChatGPT while 82% have heard little to no information on the subject. Since the AI model was the first to achieve widespread use among non technical people, not many people know how large language models (LLM) work and what they can do.

ChatGPT can be used for multiple different purposes such as creative writing, interactive learning, and answering questions. “I use ChatGPT for most of my writing,” said Ms. Baney, a teacher at Bedford High School, “It is an excellent tool for writing college recommendation letters.” It is important to keep in mind that ChatGPT is not researching or thinking about what it says. The AI is designed to be an advanced text prediction software. It uses the previous context of the conversation to guess the most likely next word. This means that ChatGPT isn’t able to distinguish fact from fiction and will often hallucinate false answers.

Unlike most programs, AI programs aren’t made by humans. ChatGPT was made by a process called training. Training is similar to evolution, to put it simply, the AI evolves to give the correct answer with the given inputs. In order to train ChatGPT OpenAI needed two main things: computing power and training data. The computing power was easy, all they needed was hundreds of computers, the hard part was getting the training data.

In order to train something like ChatGPT OpenAI needed terabytes of data. This is a massive amount of data, Wikipedia is only about 100 gigabytes. This data was collected from thousands of sources, classical literature, Wikipedia, and Reddit just to name a few. This data was used to align the models output and create what we now call GPT3. OpenAI then created a chat program that used the learned knowledge from GPT3 to get the final product of ChatGPT. An OpenAI statement about the benefits of the chat system said “The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests”.