boreal mountain
The OpenAI API has plenty of rules. Much of these seem bizarre and conflicting. Say, there's a rule that any publicly viewable use of the model output should be manually reviewed, essentially banning all possible uses where you would integrate the model output to some tool that doesn't have human review. Also, there's the content policy, which seems to apply regardless of publishing the model output or not, say, based on the rules, I cannot ask in the playground the model to generate me financial advice to see how well it does, because it goes against terms of use. The terms seem very broad in a manner that makes sense in case I were to publish the resulting completions, or integrate them onto some tool used by third parties, but from what I can tell, these apply to Playground as well. However, then there are also sections of integrating API onto automated tools without such oversight.
What brought this to mind was, I was considering having AI roleplay as video game character, and I couldn't help but notice that in a typical video game in the RPG/Action/FPS genre, you would violate the "violent content" policy pretty much by just providing setting of the game. Thinking about it, I realized I couldn't even try and check in the playground how the output would play out because it would risk model producing violent content, which seems to be expressly against terms of use.
I also had hoped to integrate GPT-4 to a smart home type system(Initially to be used just by me), which would be seemingly grey zone with "activity that violates privacy" section of allowed use. And because the terms of allowed use do not distinguish at all between crafting prompt in playground and releasing a full solution commercially to third parties, even trying out the concept in playground seems like a minefield for getting access blocked.
So I guess the question is, does the allowed use differentiate between public tool integration and playground testing?