OpenAI has expanded access to its advanced text-to-image generator, DALL-E 3, by offering it to ChatGPT Plus and Enterprise subscribers. DALL-E 3, unveiled last month, utilizes the natural language processing capabilities of ChatGPT to create images based on detailed textual prompts provided by users. The new system seeks to enhance OpenAI’s previous DALL-E 2 model, focusing on improved visual detail, crisper imagery, and responsiveness to extensive prompt descriptions.
Microsoft was the first major platform to publicly deploy DALL-E 3, integrating it with Bing Search and Bing Chat. However, there were initial concerns about problematic content getting through the system’s content filters, including images related to controversial 9/11 scenarios. OpenAI claims to have since bolstered safety measures and oversight for DALL-E 3.
OpenAI employs a multi-tiered safety system to restrict DALL·E 3’s ability to generate potentially harmful imagery, such as violent, adult, or hateful content. Safety checks are applied to user prompts and the resulting imagery before it is presented to users. Furthermore, OpenAI has implemented new measures to prevent outputs that mimic the styles of specific artists or public figures.
OpenAI is also developing an internal “provenance classifier” to identify whether an image was generated by DALL-E 3 with over 99% accuracy. Systems like DALL-E face ongoing challenges related to reproducing copyrighted content, generating nonconsensual intimate imagery, and perpetuating biases.
OpenAI intends to refine DALL-E 3’s safety further through user feedback and expert guidance. While the expansion of DALL-E 3 to ChatGPT subscribers signifies a significant increase in publicly available AI image generation capabilities, the risks associated with harmful content and intellectual property violations persist. Industry-wide collaboration on AI ethics and the establishment of reasonable regulations are expected to become increasingly important moving forward.
