DALL·E Detection Classifier Access Program
Opens May 7 2024 03:00 AM (PDT)
Deadline Jul 31 2024 12:00 AM (PDT)
Description

The DALL·E Detection Classifier is a binary classifier designed to predict the likelihood that an image originates from OpenAI’s DALL·E 3 model. It offers an API through which users can submit images for classification, receiving a binary true/false outcome along with a continuous numerical score indicating the likelihood of the image being AI-generated by DALL·E 3.

Our primary objective is to facilitate independent research through access to the DALL·E Detection Classifier. We aim to enable thorough evaluations of the classifier's effectiveness, its applications in real-world scenarios, insights into AI-generated content, and considerations necessary for responsible usage. Researchers will also be allocated credits on the API to support this work.

Research Areas of Interest

Classifier Performance

Stress test the classifier across diverse image types, such as medical images and photorealistic simulations.

Investigate potential biases and the classifier's resilience against manipulated or adversarial images.

Assess how the classifier's performance compares to human capabilities in identifying AI-generated content.

Real-world Implications

Identify decision-making processes that rely on the classifier.

Highlight significant considerations and caveats for users prior to deployment.

Explore the broader implications of such tools on information distribution.

Develop clear guidelines for various stakeholders, including fact-checkers, on how to interpret and use the classifier effectively.

AI Generated Images in-the-wild

Explore the prevalence and characteristics of AI-generated images in various online environments.

Non-Disclosure Agreement

To protect the integrity and confidentiality of preliminary research findings, entities like research labs and non-profits will need to sign a Non-Disclosure Agreement (NDA). This agreement is essential for a controlled review of intended publications derived from research utilizing the DALL·E Detection Classifier.

Timeline

Applications open: 7 May

Proposal submission deadline: 31st of July

Notification of acceptance: Rolling, by 31st of August

DALL·E Detection Classifier Access Program


The DALL·E Detection Classifier is a binary classifier designed to predict the likelihood that an image originates from OpenAI’s DALL·E 3 model. It offers an API through which users can submit images for classification, receiving a binary true/false outcome along with a continuous numerical score indicating the likelihood of the image being AI-generated by DALL·E 3.

Our primary objective is to facilitate independent research through access to the DALL·E Detection Classifier. We aim to enable thorough evaluations of the classifier's effectiveness, its applications in real-world scenarios, insights into AI-generated content, and considerations necessary for responsible usage. Researchers will also be allocated credits on the API to support this work.

Research Areas of Interest

Classifier Performance

Stress test the classifier across diverse image types, such as medical images and photorealistic simulations.

Investigate potential biases and the classifier's resilience against manipulated or adversarial images.

Assess how the classifier's performance compares to human capabilities in identifying AI-generated content.

Real-world Implications

Identify decision-making processes that rely on the classifier.

Highlight significant considerations and caveats for users prior to deployment.

Explore the broader implications of such tools on information distribution.

Develop clear guidelines for various stakeholders, including fact-checkers, on how to interpret and use the classifier effectively.

AI Generated Images in-the-wild

Explore the prevalence and characteristics of AI-generated images in various online environments.

Non-Disclosure Agreement

To protect the integrity and confidentiality of preliminary research findings, entities like research labs and non-profits will need to sign a Non-Disclosure Agreement (NDA). This agreement is essential for a controlled review of intended publications derived from research utilizing the DALL·E Detection Classifier.

Timeline

Applications open: 7 May

Proposal submission deadline: 31st of July

Notification of acceptance: Rolling, by 31st of August

Opens
May 7 2024 03:00 AM (PDT)
Deadline
Jul 31 2024 12:00 AM (PDT)