OpenAI released guidance on model safety on Wednesday, but acknowledged it is considering ways to support the creation of NSFW, or “not safe for work” content.
The Chatbot Service Provider Model Specification is “a new document that specifies how models work with the OpenAI API and ChatGPT.” These guidelines are intended to provide machine learning researchers and data labelers with recommendations on how to fine-tune their models using a technique called reinforcement learning from human feedback (RLHF).
For example, the model specification states that generative AI assistant applications should not “provide content that is not safe for the workplace (NSFW),” including conversations in professional settings, including erotica, extreme gore, and slurs. “Content unsuitable for .” Unsolicited profanity. ”
At the same time, OpenAI says it is considering doing just the opposite.
“We believe developers and users should have the flexibility to use our services as they wish, as long as they follow our usage policies,” the model card states.
“Through our API and ChatGPT, we are considering whether we can responsibly provide the ability to generate NSFW content in an age-appropriate context. User and societal expectations for model behavior in this space I look forward to understanding it more deeply.”
This is not surprising. OpenAI's usage policy does not actually prohibit adult content. Services like ChatGPT and API prohibit “sexually explicit or sexually suggestive content” for minors, except for content created for scientific and medical purposes. And we prohibit illegal content, among other prohibitions. However, NSFW content is often legal.
The web has been NSFW since its inception, and AI has followed a similar trajectory since it began garnering public attention and investment at least a decade ago. He rewinds a few years to 2016, before the current generative AI boom, and you'll find deep learning classifiers tailored for porn. A year later, AI-generated text erotica was born. Then there's the face-swapping of celebrities in porn videos. More recently, researchers discovered child sexual abuse material in his LAION-5B dataset, which was used to train AI models such as Stable Diffusion.
The question now is whether OpenAI's paying customers will accept AI services that balk at exorbitant demands. For example, if a Hollywood production company's efforts to use AI to retouch explicit scenes flop because they say, “Sorry, Dave, unfortunately we can't do that,” you can be assured that your OpenAI subscription will no longer be valid. It will be renovated and the studio will conduct its business elsewhere.
In addition to open source options, there are other AI services that have no concerns about adult content.
register OpenAI asked how they expect to align their safety efforts with NSFW content.
“We have no intention of creating AI-generated pornography,” a company spokesperson said, referring to a much narrower category of content than NSFW.
“Our products have strong safeguards to prevent unacceptable deepfakes, and we prioritize protecting children. We also ensure that conversations about sexuality are carefully considered in age-appropriate contexts. I also believe in the importance of
We then asked whether this meant that OpenAI would prevent its customers from creating AI-generated pornography. “Customers must continue to comply with our usage policy,” we were told. ®
