Hello! The AI in the playground says that:
Human: why do I have to pay for playground?
AI: There is no cost to use the playground, but you might find some features more useful if you choose to donate to OpenAI.
And support tells me that playground uses my tokens. But why. When I use playground I help the AI to improve and OPENAI can improve it with my input, so why do I have to pay?
No, you are not helping in anyway (except perhaps usage statistics, if they care), nor is the AI learning with you.
If we didnt have to pay for the playground, everyone would just use the playground for big requests. The playground is simply an easier interface to get to the OpenAI API. Makes sense that the same costs apply.
But why does the AI say it’s free forever?
valueable lesson: dont trust everything the AI says
I never trust anything AI says, but it should know if it is free or paid.
how would it know? The playground was released with GPT3, which means that during its training (before release) the playground didnt exist.
Well technically the model responded correctly, your prompt leaves space for ambiguity. Because is true you dont have to pay for the use of the playground, you have to pay for the amount of tokens you use in your completions.
On the topic of how much you are teaching the AI by using the playground, I’m under the impression ~1/100 prompts are used to improve the AI at this time.
Last I saw, 0% of the prompts are used to teach the AI
I would be curious to know which of you (@MChaney22 or @TimC) are correct. I took a year hiatus from the api after spending tons of time on it during my trial period. Now that I have to pay I only use it when I actually have some reason worth putting money behind. That said, when I was using it a little over a year ago, my understanding was that it was 100% static environment post training; I didn’t think public interaction was used as some kind of ongoing training, but if it it, I would be interested to know.
I found the policy on the help blog updated 1 week ago. How your data is used to improve model performance | OpenAI Help Center
Current policy seems be that they’ll use at most 200 prompts per user per 6 months. They filter out personally identifiable info. You can also opt out of this by emailing firstname.lastname@example.org
Ah interesting, makes sense that they are using that data now that they have a lot of it. Seems like this is a source for targeted investigations rather than feeding prompts directly going back to the model.