DALL·E 2 pre-training mitigations
OpenAI Blog
OpenAI implemented content policy guardrails into DALL·E 2's training process to mitigate risks before making the image generation model widely available. The company applied pre-training mitigations but did not specify measurable benchmarks or quantified risk reductions in this announcement. The guardrails aim to prevent the model from generating images that violate OpenAI's content policy when users interact with the deployed system.
Why it matters
In order to share the magic of DALL·E 2 with a broad audience, we needed to reduce the risks associated with powerful image generation models. To this end, we put various guardrails in place to prevent generated images from violating our content policy.