HACKER Q&A
📣 MillionOClock

Finetuning OpenAI models and ToS restrictions


I think I read somewhere that OpenAI didn't allow using its models output to train other models, does this also apply to their own fine tuning API?

  👤 ilaksh Accepted Answer ✓
I had the same question. I feel like they should clarify this. But also I don't see how they would want to apply that restriction in this case, since using their models to generate training data will make them more money and makes fine-tuning vastly more manageable.
👤 tikkun
(Not the OP)

Specifically the TOS says:

“You may not … (iii) use output from the Services to develop models that compete with OpenAI;”

It would be interesting to clarify:

1) does this apply to the fine tuning api (probably yes)

2) what counts as competing with OpenAI (probably just selling general non-use case specific llm apis and platforms? I’d hope that vertical specific apis are ok and chatgpt competitors are ok)

3) does creating a model with openai’s fine tuning engine itself count as creating a model or are they intending this to apply to non-OpenAI hosted model creation only (I’d hope that they’re fine with people creating fine tuned models that compete with OpenAI given it’s on their platform)