On the last piece for content produced by general AI tools. This is also something so I think Anita already mentioned this, I think Max is also doing this. So for our DALL·E image generation technology we’ve also added the C2PA protocol, that coalition for content provenance and authenticity, which is basically some metadata that’s inserted into the narratives that are generated so that you can see what tool generates. So for this, that’d be DALL·E. I will say that this is not foolproof by any means people could screenshot the image. Or they could delete the metadata at times when it’s uploaded to certain platforms to probably reduce the file size, and the data may also be altered or removed. So I think this is one tool and toolkit but not by no means the full solution. We’ve also been exploring the provenance classifier, so a tool that will allow us to detect if images were created by DALL·E even without that C2PA watermark.

Keyboard shortcuts

j previous speech k next speech