Before GPT-3, language models like GPT-2 showed surprising versatility—translation, summarization, and question answering emerged purely from next-word prediction. However, they still struggled to reliably adapt without task-specific fine-tuning. Prompts had to be carefully crafted, and real-world applications often required retraining. GPT-3 tackled a bolder question: what if we scale a language model to an extreme size, with 175 billion parameters? The result transformed AI. GPT-3 demonstrated that with enough scale, models could learn new tasks from just a few examples in the prompt—no gradient updates needed. This capability, known as few-shot or in-context learning, became the foundation for modern systems like ChatGPT. Below, we answer key questions about this landmark paper.
