GPT-3 is a 2020 OpenAI model, 175 billion parameters in size, which showed a big model could do new tasks from a few examples in the prompt.
In plain words
Earlier models usually needed extra training for each task. GPT-3 was far bigger than any previous dense language model, and it could often pick up a task just from instructions and a handful of examples written into the prompt, with no retraining. This is called few-shot learning. It still struggled on some tasks, and people found its fake news articles hard to tell from real ones.
A quick example
A few-shot prompt might list three English words with their French translations, then a fourth English word with a blank. GPT-3 fills in the French, having learned the pattern only from the prompt.
Where you'll meet it
- Research on few-shot prompting and in-context learning.
- Tasks such as translation and question answering from a prompt.
Related
Sources
- PaperLanguage Models are Few-Shot Learners, Brown et al. (arXiv) · read 29 Sept 2026
- DocsLanguage models are few-shot learners, OpenAI · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.