Carl T. Bergstrom on Nostr: If openAI were digitally watermarking the output of ChatGPT, so they could exclude it ...
If openAI were digitally watermarking the output of ChatGPT, so they could exclude it from future training data, but not telling anyone so as to have a competitive edge, would there be a way to detect that?
Published at
2024-05-27 15:18:41Event JSON
{
"id": "858b7606739cdc04f9bff68b79f7a767362afc3965a9a8ec98a69e8c2900d307",
"pubkey": "ac5404833d7aff6cebf4afd632d367ccf064470e9ad728b47bce8be6bfbb958a",
"created_at": 1716823121,
"kind": 1,
"tags": [
[
"proxy",
"https://fediscience.org/users/ct_bergstrom/statuses/112513720089145560",
"activitypub"
]
],
"content": "If openAI were digitally watermarking the output of ChatGPT, so they could exclude it from future training data, but not telling anyone so as to have a competitive edge, would there be a way to detect that?",
"sig": "42cb46537ada2a331aec546cf1cea89136779b4241bcc6cd9493e9473b8854b07f2ca4460330a272bec19639c0656a9971bb51a253a221508f1f80fa03800e96"
}