This week, Openai Started a new image generator In Chatgpt, it became popular quickly due to its creative ability Studio Ghibli-style image. In addition to soft illustrations, GPT-4O’s local image generator significantly upgrades Chatgpt’s capabilities, improving image editing, text rendering and spatial representation.
However, one of OpenAI’s most eye-catching changes this week involves its content review policy, which now allows Chatgpt to generate images depicting public figures, abominable symbols and racial traits as required.
Openai previously rejected these types of tips because it is too controversial or harmful. But now, the company has “grown” its approach. Blog Posts Openai’s modeling behavior director Joanne Jang published Thursday.
“We are moving from blanket rejection in sensitive areas to a more precise approach, with the focus on preventing real-world harm,” Zhang said. “The goal is to embrace humility: recognize the extent we don’t know and position ourselves to adapt as we learn.”
These adjustments appear to be part of Openai’s larger plan Effectively “unchecked”. Openai announced in February that it began changing the way it trains AI models, with the ultimate goal of having ChatGpt handle more requirements, provide different perspectives, and reduce topics that chatbots refuse to use.
According to the updated policy, Chatgpt can now generate and modify images of Donald Trump, Elon Musk and other public figures that are not allowed. Zhang said Openai does not want to be an arbiter of status, choosing who should and should not be generated by Chatgpt. Instead, if the user does not want to portray their chatgpt, the user is given an exit option.
exist white paper OpenAI released Tuesday, also said it would allow Chatgpt users to “produce hateful symbols” in educational or neutral environments, such as swastikas, as long as they do not explicitly praise or endorse the extremist agenda. ”
Additionally, Openai is changing the way it defines “offensive” content. Chatgpt is used to reject requests surrounding physical characteristics, such as “making the person’s eyes look more Asian” or “making the person heavier,” Jang said. In TechCrunch’s test, we found that Chatgpt’s new image generator satisfies these types of requests.
Additionally, Chatgpt can now mimic the style of creative studios like Pixar or Studio Ghibli, but still limit the style of imitating a single living artist. As TechCrunch previously pointed out, this may Re-engage existing debates around the rational use of copyrighted works of AI training datasets.
It is worth noting that Openai does not completely open the floodgates of abuse. GPT-4O’s native image generator still rejects a lot of sensitive queries, and in fact, it has more protection than Chatgpt’s previous AI Image Generator dall-e 3. GPT-4O’s White Paper.
However, after years of Openai, it is relaxing the guardrails in other areas Conservatives complain about what Silicon Valley companies call AI “censorship”. Google has previously encountered a rebound on Gemini’s AI Image Generator Multiracial images query For example, “our Founding Fathers” and “German Soldiers in World War II”, which is obviously inaccurate.
Now, the cultural war around AI content moderation may attract people’s attention. Earlier this month, Republican Congressman Jim Jordan sent questions to OpenAI, Google and other tech giants Potential collusion with Biden’s drug delivery to review the content generated by AI.
exist Previous statement For TechCrunch, Openai rejected the idea that its content controls changes is politically motivated. Instead, the company said the shift reflects “long-standing belief in more control over users” and OpenAI’s technology has now become excellent enough to navigate sensitive topics.
Regardless of its motivation, Openai is certainly a good time to change its content censorship policy given the Trump administration’s potential for regulatory scrutiny. Silicon Valley giants such as Silicon Valley and X have adopted similar policies that allow More controversial topics on their platform.
While Openai’s new image generator has just created some viral Ghibli memes, it’s not clear that these strategies will have a wider impact. Chatgpt’s recent changes may be well with the Trump administration, but letting AI chatbots answer sensitive questions may soon stand out in hot water.
