massive neural networks for text. Play Large Language Models trivia solo to sharpen your knowledge, or challenge a friend head-to-head in Trivia Tango — every question comes with an explanation so you learn as you play.
Think you know the answers? Play to find out.
The "T" in GPT stands for this neural network architecture introduced in 2017 that revolutionized natural language processing through self-attention mechanisms.
When you type a message to an AI assistant, each word or word-piece is converted into one of these basic units that the model actually processes.
This technique uses human ratings of AI responses to train a reward model, which then guides the AI to produce more helpful and safer outputs.
This company, founded in 2021 by former members of the ChatGPT team, created Claude and focuses on AI safety research.
The maximum amount of text a model can process at once, measured in tokens, is called this—and it determines how much conversation history the model can "remember."
This phenomenon occurs when a language model confidently generates false or fabricated information that sounds plausible but has no basis in reality.
These numerical representations convert words into dense vectors, capturing semantic relationships so that "king" and "queen" are mathematically closer than "king" and "banana."
This training objective, used by GPT models, involves predicting the next word in a sequence given all the previous words.