As communicated on the official blog of OpenAI, ChatGPT is now able to create images from a common conversation, through the integration of FROM AND 3.
This new feature, available for Plus users ed Enterprisewill therefore allow us to create a synergy between two of the most appreciated tools in the context of generative Artificial Intelligence.
At present, DALL·E 3 represents the ultimate in image generation from text prompts. Compared to its predecessor, in fact, this tool is able to create not only visually appreciable images, but also with greater clarity in the details, even in the case of reproduction of hands and facesa real Achilles’ heel for other similar services.
The platform is equipped with a advanced security system, which can prevent the creation of violent, adult or hateful content. To support these mechanisms, then, there is the processing of user feedback, which is invaluable for evaluating borderline cases.
Precisely in this context, the merger between ChatGPT and DALL·E 3 can offer new opportunities. THE feedbackas well as sensitive cases, are destined to increase and, thanks to the collaboration of users, this combination is becoming increasingly effective and at the same time safe.
DALL·E 3 and ChatGPT together: a system for “tagging” AI-generated content
OpenAI, in the same blog post, also talked about the possible introduction of a “provenance classifier“.
The reference is to an instrument that should be able to guarantee the provenance of an image, whether this was generated by DALL·E 3 or not. The instrument, put to the test with the first tests, demonstrated encouraging data.
In the 99% in fact, in some cases this has proven capable of identifying an image created by DALL·E. Not only that: even if the processing was subject to subsequent changes (clippings, downsizing o JPEG compressions), the classifier maintains an accuracy higher than 95%. In fact, it is an initiative that appears even more in-depth than what was proposed by Google just over a month ago.
This type of system for evaluating the origin of content, however, opens up new scenarios. In fact, we are talking about applications not only limited to images but also linked to audio and/or video processing. In this sense, it will be possible to combat phenomena such as i deep fake.