OpenAI's Content Moderation Policy Takes a Sharp Turn! ChatGPT's New Image Generator Goes Viral, Easing Restrictions on Controversial Content
This week, OpenAI launched a new image generator within ChatGPT, quickly gaining popularity for its ability to generate images in the style of Studio Ghibli. Powered by GPT-4o, this generator significantly enhances ChatGPT's image editing, text rendering, and spatial representation capabilities.
However, even more noteworthy is OpenAI's significant adjustment to its content moderation policy. The new policy allows ChatGPT to generate images depicting public figures, hate symbols, and racial characteristics upon user request, a stark contrast to its previous strict limitations on controversial content.
Joanne Jang, OpenAI's head of model behavior, stated that the company is shifting from a policy of "blanket rejection" to a "more nuanced approach," aiming to prevent real-world harm while emphasizing "adapting as we learn." This move is seen as part of OpenAI's larger plan to "de-moderate" ChatGPT, enabling the model to handle a wider range of requests and offer more diverse perspectives.
Under the new rules, ChatGPT can now generate and modify images of public figures like Donald Trump and Elon Musk. OpenAI believes it shouldn't act as the "arbiter" of who can be generated and provides users with an opt-out mechanism. Furthermore, in educational or neutral contexts, ChatGPT can even generate "hate symbols" such as Nazi insignia, provided there's no explicit praise or support for extremism. Restrictions have also been eased on requests previously rejected due to their depiction of physical characteristics. Simultaneously, ChatGPT can mimic the styles of studios like Pixar or Ghibli, but still restricts mimicking living artists.
Despite this, OpenAI isn't completely removing safeguards against abuse; GPT-4o has stricter protections in place for generating images of children. However, this significant change in content moderation policy comes at a time when conservative critics have long accused Silicon Valley companies of "censorship" in AI. Previously, Google's Gemini image generator sparked controversy for generating inaccurate images of multiracial historical figures.
Republican Congressman Jim Jordan has already questioned tech giants like OpenAI regarding AI content moderation. OpenAI denies that the policy change is politically motivated, claiming it reflects a "long-held belief in giving users more control" and that the technology is mature enough to handle sensitive topics.
Considering potential future regulatory scrutiny, OpenAI's timing is sensitive. Platforms like Meta and X have adopted similar strategies, allowing more controversial content. Although ChatGPT's new image generator is currently primarily used for creating entertaining Ghibli-style images, its shift in content moderation policy could have broader implications and potentially present OpenAI with new challenges in responding to sensitive issues.