ChatGPT’s Dall-E 3, a fascinating example of AI artistry, has recently been unveiled, providing a glimpse into the capabilities of artificial intelligence. However, concerns about the quality and ethical implications of this technology also arise.
OpenAI has implemented a usage cap of 50 prompts every three hours for ChatGPT’s Dall-E 4 version due to the high computational power required to process these prompts. Users who exceed this limit will encounter an error message indicating the duration of the waiting period before regaining access. Additionally, there might be a slight delay of around 30 seconds before receiving the generated images. OpenAI’s guidelines strictly regulate the types of images that can be produced, and any creations that violate these guidelines may be withheld or replaced with compliant images.
One notable difference with ChatGPT’s Dall-E 3 compared to previous AI image generators like Dall-E 2 or Midjourney is the addition of ChatGPT as an intermediary. ChatGPT generates diverse prompts, varying from long sentences to complete paragraphs, each tailored to facilitate Dall-E 3’s image creation. For instance, when requesting an image of two WIRED reporters interviewing a CEO, ChatGPT would provide a detailed prompt mentioning the gender, race, and roles of each individual within the image.
Notably, artists have expressed concern over the lack of protection in this new update. While the chatbot refuses to create images mimicking contemporary artists, there are still potential workarounds. For example, when asked to design a coffee mug inspired by Keith Haring’s style, the AI tool declined but offered a compromise by creating a design inspired by general characteristics such as bold lines, vibrant colors, and simplistic figures. However, the resulting designs were often subpar and messy.
Although the quality of the art generated by Dall-E 3 has improved, there are still underlying issues with image generators. The images may appear human-made at first glance, but upon closer inspection, anomalies and uncanny details become evident. Some comical errors may occur, such as the chatbot struggling to label baking ingredients correctly. However, more concerning mistakes are also present. For instance, when asked to create a map of Israel and the Gaza Strip, ChatGPT repeatedly mislabeled Gaza as part of the Mediterranean Sea.
Another challenge these image generators face is their tendency to rely on racist stereotypes when depicting humans. Dall-E 3 is no exception. Out of the 20 images requested to depict WIRED reporters, the chatbot primarily produced white and male representations, deviating from the intended diverse portrayal except in a few instances. Without specific prompts for race and gender, the results defaulted to biased depictions.
Although Dall-E 3 showcases remarkable strides in AI-generated art, concerns regarding quality and ethical implications persist. Efforts to improve upon these issues are necessary to ensure the responsible use of AI technology.
