This week, Opeli launched a up-to-date painting generator in ChatgPT, which quickly became viral because of his ability to create images in the style of Studio Ghibli. In addition to pastel illustrations, the native GPT-4O image generator significantly improves CHATGPT, improving image editing, text rendering and spatial representation.
However, one of the most noteworthy OpenAI changes this week includes the rules regarding the moderation of content that now allow chatgPT, on demand, generation of images depicting public numbers, hateful symbols and racial features.
OpenAi previously rejected this kind of hints that he is too controversial or harmful. But now the company has “evolved” its approach, as Blog post Published Thursday by the Behavior Lead model, Joanne Jang.
“We are moving from the vocabulary in sensitive areas to a more precise approach focused on preventing damage to the real world,” said Jang. “The goal is to accept humility: recognizing how much we do not know, and setting up to adapt when we learn.”
These corrections seem to be part of a larger OpenAI plan to effectively “senselessly” “” “” without “chatgpt. Opeli announced in February that he was starting to change the way he trains AI models, with the final goal, which is to enable chatgpt, offering various perspectives and reduce topics with which chatbot refuses to work.
According to the updated principle, CHATGPT can now generate and modify the paintings of Donald Trump, Elon Musk and other public figures, which Openai has not previously allowed. Jang says that Opeli does not want to be a status arbiter, choosing who should and should not be generated by ChatgPT. Instead, the company gives users the option to resign if they do not want ChatgPT to present them.
IN White paper Published on Tuesday, Opennai also said that it would allow chatgpt users to “generate hateful symbols”, such as swastikas, in educational or neutral contexts, unless they “clearly praise or support extremist programs.”
In addition, OpenAI changes the way it defines “offensive” content. Jang says that ChatgPT rejects requests for physical features, such as “make the person’s eyes look more Asian” or “make this person harder.” In TechCrunch tests, we found that the up-to-date CHATGPT image generator fulfills this kind of demand.
In addition, chatgpt can now imitate artistic styles – such as Pixar or Studio Ghibli – but still limits the imitation of individual artists. As TechCrunch noted earlier, this can reproduce the existing debate on the permitted operate of copyrights protected by copyright in AI training sets.
It is worth noting that OpenAI does not completely open the flood gates for improper operate. The native GPT-4O image generator still refuses many sensitive queries, and in reality it has more security around generating children’s images than Dall-E 3, the previous AI ChatgPT image generator, according to the AI ChatgPT image generator, White Book of GPT-4O.
But OpenAi relaxes handrails in other areas after years of conservative complaints around the alleged “censorship” AI from companies from the Silicon Valley. Google has previously encountered the reaction of the AI Gemini image generator, which created multi -track paintings for queries, such as “Ujców founders of the USA” and “German soldiers in World War II”, which were obviously wrong.
Now the cultural war around the AI content moderation can come to mind. At the beginning of this month, the Republican Congressmen Jim Jordan sent questions to OpenAI, Google and other technological giants about potential collusion with Biden administration in order to censor the content generated by AI.
In the previous statement for TechCrunch, Opennai rejected the idea that the changes in content moderation were politically motivated. Instead, the company claims that the change reflects “the belief in providing users for a long time”, and OpenAi technology is now good enough to move with sensitive objects.
Regardless of its motivation, it is certainly a good time for Otnai to change the rules of content moderation, taking into account the potential of regulatory control within the Trump administration. Giants from the Silicon Valley, such as Meta and X, also adopted similar policies, enabling more controversial topics on their platforms.
While the up-to-date OPENENAI image generator has created only some of the Ghibli memes so far, it is not clear what the wider effects of these rules will be. Recent ChatgPT changes can come to terms with Trump’s administration, but allowing sensitive questions in chatbot AI may soon land in heated water.