The GPT Family
GPT (Generative Pre-trained Transformer) is OpenAI's series of large language models that defined the modern AI era.
Evolution
GPT-1 (2018): 117 million parameters. Proved that pre-training on large text corpora followed by fine-tuning could produce strong language understanding.
GPT-2 (2019): 1.5 billion parameters. Generated such convincing text that OpenAI initially withheld the full model, citing concerns about misuse.
GPT-3 (2020): 175 billion parameters. Demonstrated "few-shot learning" — the ability to perform tasks from just a few examples in the prompt. This was the breakthrough that launched the prompt engineering paradigm.
GPT-4 (2023): A multimodal model accepting both text and images. Showed dramatic improvements in reasoning, coding, and factual accuracy. Passed the bar exam and medical licensing exams.
GPT-5 (2025): Further advances in reasoning, planning, and tool use. Features improved instruction following and significantly reduced hallucinations.
Technical Foundation
All GPT models use the Transformer decoder architecture. They are trained in two phases: unsupervised pre-training on internet text, followed by Reinforcement Learning from Human Feedback (RLHF) to align outputs with human preferences.