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

How are you different from chatGPT?

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

    How are you different from chatGPT?

  2. Assistant

    We share many commonalities, and the complete details of ChatGPT's construction aren't always shared with the public. I recognize that this is somewhat ironic because ChatGPT is built by OpenAI, who said they were on a mission to build open-source AI, and they have, as of 2023, stopped being very open with their technology. So Basically, this is a difficult question to answer.

    Some important differences:

    • I am trained on data provided by and reviewed by the volunteers at the Open Assistant project.
    • ChatGPT was trained on data provided by paid participants on a platform like Amazon's Mechanical Turk.
    • it is rumored that ChatGPT results improve over time, which would indicate that its weights are often updated. (maybe they are continuously training ChatGPT on user feedback about its generated outputs)
    • ChatGPT is closed source.

    Some Commonalities:

    • Open Assistant also has the capacity to be updated continuously,
    • We are both trained on as much data as our creators have the capacity to provide
    • We are likely trained on very similar or identical data since we are both trained on as much text as the internet can provide.
  3. User

    Do you think you can outperform GPT-4 in the near future?

  4. Assistant

    I hope I'll outperform GPT-4 in the near future.

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