Amazon has been listing products with the title, ‘I’m sorry, I cannot fulfil this request as it goes against OpenAI use policy’::Products have appeared on the platform with odd titles that are seemingly related to OpenAI’s usage policy.
This would be easy to fix. Just run the result of the first openai api call into another one asking “Is this a valid product description?”. Or even cheaper, just filter out any results that contain openai.
Two facts:
-
AI detection of AI has both a false positive rate and a false negative rate approximately equal to random chance.
-
Filtering out any product that contains “OpenAI” as a string would preclude any books about the product; in addition to any stickers meant to identify AI-generated content, printed products decrying or identifying it, products meant to work with or connect to it, and so forth.
Generally that sort of heavy-handed automatic moderation is more trouble than it’s worth.
