Open AI

Source: OpenAI Documentation

Introduction

• OpenAI API can be use for any task involved in understand/generating natural language or code.
• Different models are available for different tasks depending on power and task:
  • Content generation
  • Summarization
  • Classification, categorization, sentiment analysis
  • Data extraction
  • Translation
  • Conversation
  • Creative writing
  • Style changes
  • And More

Tutorial

• Create good instructions to get good results
  • Tell the model what you want
  • Give examples of what you want

Suggest three names for an animal that is a superhero.

Animal: Cat
Names: Captain Sharpclaw, Agent Fluffball, The Incredible Feline
Animal: Dog
Names: Ruff the Protector, Wonder Canine, Sir Barks-a-Lot
Animal: Horse
Names: The Mighty Steed, The Great Equine, The Valiant Charger

• With a temperature above 0 (like 1), different names will be suggested for completions. Low temperature is less risky and more accurate and deterministics, while higher temperature is more risky and more creative and diverse completions.
  • For the name example, a moderate temperature of 0.6 works.
• Want to provide more examples or a better trained model? Use the fine-tuning API to provide 100+ - 1000s+ examples to customize a model for your specific use case.

Prompts and Completions

• Completions endpoint is a simple use of the API with a /prompt/ and the model generates a compleition.
• See Examples for sample prompts, how to them them, and example completions.

Tokens

• Tokens can be words or chunks of characters, generally 1 token = 4 characters / 0.75 words of English
• Most models are limited to 2048 token inputs

Models

• The API uses a set of models with different capabilities and costs.
• Base GPT-3 models are: Davinci, Curie, Babbage, and Ada. Codex series is a descendant of GPT-3 trained on natural language and code.