Dana Fried on Nostr: If you're worried that your images have already been scraped for training, be aware: ...
If you're worried that your images have already been scraped for training, be aware:
Training sets tend to be metadata-and-URL-only; nobody actually packages and distributes terabytes of (often copyrighted!) image data,
So, if you replace existing images with nightshade+glaze, any new models will receive the altered payload.
Published at
2024-01-19 19:05:12Event JSON
{
"id": "7b8501073e0dc8f6149d89c0d92e8e80f7820c6ae85a2edd65ffbc6a2a0be2f9",
"pubkey": "8139cfb427b06d70923c467ab1eb5c52f375c493f9efa24d8d63c5e23b64f628",
"created_at": 1705691112,
"kind": 1,
"tags": [
[
"e",
"823e014a5c45dec9bc90cec6171755f4fc38e6084e8603d0eacda0dc60976416",
"wss://relay.mostr.pub",
"reply"
],
[
"proxy",
"https://mastodon.social/users/tess/statuses/111784172717076139",
"activitypub"
]
],
"content": "If you're worried that your images have already been scraped for training, be aware:\n\nTraining sets tend to be metadata-and-URL-only; nobody actually packages and distributes terabytes of (often copyrighted!) image data,\n\nSo, if you replace existing images with nightshade+glaze, any new models will receive the altered payload.",
"sig": "77548e3baaf5e7657975fc23e05551f554dc0082720c9c17afdf3384c438c19ffdd998eb823d5b492f4d56ce768ce8a9df48150dac7deef620e99214db018efb"
}
