Skip to content
TheChatArchiveThe Chat Archive
Human-written demonstration (OpenAssistant volunteers)· OpenAssistant (human-written)Hugging Face OpenAssistant Conversations (oasst2)

What differences are there between ChatGPT and Instructional GPT?

In OpenAssistant data, “assistant” turns were written by human volunteers as demonstrations. They are not model outputs.
  1. User

    What differences are there between ChatGPT and Instructional GPT?

  2. Assistant

    Both ChatGPT and InstructGPT are auto-regressive large language models trained and made available by OpenAI. They are sibling models in the GPT-3.5 family and have been pretrained on natural language and code. They have also both been fine-tuned on human instructions and feedback.

    The ChatGPT model and interface is optimized for use as a dialog agent by human users. It is available for free on a somewhat limited basis or can be purchased for a monthly fee. It has been trained specifically as a dialog agent and has enhanced safety measures to allow it to be exposed directly to end users.

    The InstructGPT model is similar, but not identical, to the model that powers ChatGPT. It is not as restrictive as the ChatGPT model in terms of safety. It is not easily accessible to the public and is instead made available as a paid API. Requests to InstructGPT are billed based on the number of tokens returned, whereas ChatGPT is billed at a fixed monthly price regardless of usage.

    Overall, these are very similar models that can be used for similar purposes, but ChatGPT is best suited for dialog with end users, while InstructGPT is useful for developers integrating API calls to a large language model into other applications.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.