GPT-3

GPT-3 is the third-generation Generative Pre-trained Transformer from OpenAI, introduced in the May 2020 paper 'Language Models are Few-Shot Learners'. With 175 billion parameters trained on a filtered mix of Common Crawl, WebText, books, and Wikipedia, it was roughly 100 times larger than GPT-2 and demonstrated that scale alone could produce startling new behavior.

Its defining contribution was few-shot learning: instead of fine-tuning, users could give the model a few examples in the prompt itself and it would generalize — the birth of prompt engineering. GPT-3 also marked the pivot to closed, API-only releases; the weights were never published, and access came exclusively through OpenAI's paid API.

Though superseded by later models, GPT-3 is a genuine landmark: it set the scaling-laws agenda, launched the commercial LLM API era, and its InstructGPT-tuned descendants led directly to ChatGPT.