The reluctance to release the tool may be linked to the controversy surrounding OpenAI's previous public classifier tool, which was criticized for its low accuracy rate. The company is considering whether the image classifier should also detect images created with other non-OpenAI generative tools. However, the decision will depend on the reception of the image classifier tool in its current form.
Key takeaways:
- OpenAI is debating the release of a tool that can determine whether an image was made with DALL-E 3, its generative AI art model. The tool's accuracy is considered "really good" but hasn't met OpenAI's quality threshold yet.
- The tool's targeted accuracy is extraordinarily high, with a reported reliability of 99% at determining if an unmodified photo was generated using DALL-E 3. However, OpenAI is also grappling with the philosophical question of what constitutes an AI-generated image.
- OpenAI's hesitation may be linked to the controversy surrounding its previous public classifier tool, which was designed to detect AI-generated text. The tool was pulled due to its low accuracy rate.
- Several organizations, not just OpenAI, are exploring watermarking and detection techniques for generative media as AI deepfakes become more common. However, the industry has yet to agree on a single watermarking or detection standard.