OpenAI’s latest text-to-image generator model ‘DALL-E 3’
that was first announced last month is now being launched to ChatGPT Plus and
Enterprise customers. The technology will be available within the ChatGPT app,
and is designed to provide more safety tools and the ability to enter longer
prompts.
DALL-E 3 is a more efficient version of DALL-E 2, as it enables users to produce longer and more visually descriptive prompts that the bot can then use to generate images. DALL-E 3 is also available within the Bing Chat and Bing Image Generator, which means that Microsoft was the first platform to introduce a wider public access to the model, before ChatGPT could integrate it.
As with any other AI technology, DALL-E has also faced criticism
over its failure to tackle problems related to copyright, nonconsensual nudes,
shifted ethnicity of subjects, and photo-realistic misrepresentations of public
figures. Another reason that DALL-E 2 was criticised was that it often
ignored specific wording. And therefore, OpenAI has developed DALL-E 3 to understand
context in an improved manner.
OpenAI claims, however, that DALL-E 3 will take better care
of such potential issues and harmful results, as it has been trained to limit the
generation of content in the style of living artists and images of public
figures. DALL-E 3, according to OpenAI, will also work to “improve demographic
representation across generated images.”
Additionally, OpenAI has launched an internal ‘provenance classifier’ tool that can detect, with 99% accuracy, if an image was generated by DALL-E 3.
