What is ChatGPT?
ChatGPT is a chatbot that runs on top of the GPT-3 (Generative Pre-trained Transformer) family of language models, built by OpenAI. GPT-3 is a prediction model that uses deep learning to produce text.
ChatGPT uses these language models to interact with the user in a conversational way. In addition to these language models, it was further trained using Reinforcement Learning from Human Feedback (RLHF) - meaning that human AI trainers provided example conversations and ranked AI responses.
ChatGPT can follow a conversation, write and debug computer programs, play games, write essays, and more. It can also remember and reference previous prompts given by the user, which is a feature that many other chatbots do not have.
ChatGPT has limitations. It is limited to knowledge pulled from its training data and can sometimes produce responses that sound correct but are false or nonsensical. Trainers tended to prefer longer answers and it suffers from common over-optimization issues (see Goodhart's law), so it has a tendency to be overly wordy and overuse certain phrases. There are also risks of algorithmic bias and misuse that appear in many AI implementations.