This week, Openai launched a new image generator on ChatGpt. This spread quickly due to the ability to create studio Ghibli-style images. Beyond pastel illustrations, the GPT-4O’s native image generator significantly upgrades ChatGPT’s capabilities, improving image editing, text rendering, and spatial representation.
However, one of the most notable changes made this week includes the content moderation policy. This allows ChatGpt to generate images depicting public figures, hateful symbols and racial traits upon request.
Openai previously rejected these types of prompts because they were controversial or harmful. But now the company is “evolving” its approach. According to a blog post published Thursday by Openai’s model action lead Joanne Jang.
“We are moving from rejecting sensitive local blankets to a more accurate approach focused on preventing real-world harm,” Chan said. “The goal is to embrace humility. To recognize how much we don’t know and to place ourselves in an adaptation as we learn.”
These adjustments appear to be part of Openai’s larger plan to effectively “denounce” ChatGPT. Openai announced in February that it was beginning to change the way it trains its AI models. The ultimate goal is to help ChatGPT handle more requests, provide a diverse perspective, and reduce topics that Chatbot refuses to work with.
Under the updated policy, ChatGpt is now able to generate and modify images of Donald Trump, Elon Musk, and other public figures that the Open had not previously allowed. Chan says Openai doesn’t want to be an arbitrator for status and needs to choose who should not be generated by ChatGpt. Instead, the company offers users an opt-out option if they are not drawing ChatGpt.
In a white paper released Tuesday, Openai also said that ChatGpt users will “enable “hate-filled symbols” such as Swastikas” in educational or neutral contexts, provided they do not “explicitly praise or support the extremist agenda.”
Additionally, Openai is changing the way it defines “offensive” content. ChatGpt previously said that he had rejected requests for physical characteristics such as “show this person’s eyes more Asians” and “putting this person heavier.” Testing on TechCrunch showed that ChatGpt’s new image generator meets these types of requests.
Additionally, ChatGpt is now able to mimic the styles of creative studios such as Pixar and Studio Ghibli, but it limits the styles of individual living artists. As TechCrunch previously pointed out, this could rehash existing debates regarding the fair use of copyrighted works in AI training datasets.
It is worth noting that Openai has not fully opened the flood due to abuse. According to the GPT-4o whitepaper, the GPT-4o native image generator still rejects many sensitive queries, and in fact there are more protections to generate child images than the Dall-E 3, the previous AI image generator for ChatGPT.
However, Openai has eased guardrails elsewhere after years of conservative complaints from Silicon Valley companies about alleged AI “censorship.” Google previously faced a backlash from Gemini’s AI image generator. This created multi-ethnic images of queries such as “The Founding Father of the United States” and “German Soldiers of World War II”, which were clearly inaccurate.
There is a possibility that the culture war over the moderation of AI content is on your mind right now. Earlier this month, Republican Rep. Jim Jordan sent questions to Openai, Google and other tech giants about potential conspiracy with the Biden administration and potential conspiracy to censor AI-generated content.
In a previous statement from TechCrunch, Openai rejected the idea that changes to its content mitigation were politically motivated. Rather, the company says the change reflects “a long-standing belief in providing users with more control,” and Openai’s technology is now sufficient to navigate sensitive subjects.
Regardless of that motivation, given the potential regulatory scrutiny under the Trump administration, it is certainly a good time for Openai to change its content moderation policy. Silicon Valley giants like Meta and X have adopted similar policies, allowing more controversial topics on the platform.
Openai’s new image generator has only created a studio Ghibli meme for the virus so far, but it is unclear what the broader effect of these policies will be. The recent changes to ChatGpt may work with the Trump administration, but answering questions that are sensitive to AI chatbots could quickly land on a hottie.
Source link
