GPT-3 is a GPT model introduced in 2020. Its largest version has 175 billion parameters.

Architecture and training

  • Uses a decoder-only Transformer
  • Learns left-to-right next-token prediction
  • Trains on a large text mixture that includes filtered Common Crawl data

In-context learning

GPT-3 showed strong In-Context Learning:

SettingInformation in the promptWeight update during the task
Zero-shotTask descriptionNo
One-shotTask description and one exampleNo
Few-shotTask description and several examplesNo

A prompt can demonstrate patterns such as simple arithmetic, spelling correction, or translation. The model uses those patterns to produce the next answer.

For some tasks, zero-shot or few-shot performance is comparable to task-specific fine-tuned models. This result depends on the task.

Connection to instruction following

InstructGPT fine-tunes GPT-3 using human demonstrations and RLHF. ChatGPT extends this type of training to dialogue.