How does ChatGPT work?
ChatGPT stands for "Chat Generative Pre-trained Transformer." Two models of generative pre-trained transformers (GPTs) have been released to the general public: GPT 3.5 and 4.0. Each model understands and generates human-like answers to text prompts and can interact or "chat" with people.
ChatGPT is based off hundreds of billions of words (570GB of text) that were taken from books, articles, websites, media transcripts, and social media. It uses complex algorithms known as "transformers" to place words in order in response to a prompt. The transformers are similar to rules that structure written and spoken languages.
As prompts are typed into the ChatGPT interface, words from your query are identified and converted to numerical "tokens" that are in turn passed through a pre-trained model to anticipate the next word in a sequence of words. In fact, suggested words are often provided by AI platforms as you are typing. When you submit your prompt, ChatGPT returns related information based the unique words that it identified.
As of 2024, Microsoft's free version AI, known as Copilot, currently utilizes the GPT-4 model. A subscription is required to access the premium tools and services that are offered through the Copilot interface, including text generation (based off of several platforms), image generation (based on DallE3), and video generation (based on OpenAI's Sora model). Google Gemini offers very similar services as Copilot. In addition, Copilot and Gemini have access to the Internet, so it can provide up-to-date responses on current events, something that ChatGPT does not offer.