This makes me think of cannibalism, and data prions:
“It suggests that a training diet of AI-generated text, even in small quantities, eventually becomes “poisonous” to the model being trained. Currently there are few obvious antidotes."
https://www.scientificamerican.com/article/ai-generated-data-can-poison-future-ai-models/
Lili Saintcrow /
npub1vl…pu5dt
2023-07-29 02:48:27
Author Public Key
npub1vlc9e5hvtawwxqadudvhdy8zcdynxxgqxqe452dyktzre54v7whqvpu5dtPublished at
2023-07-29 02:48:27Event JSON
{
"id": "0db0db12f66fbbef5ae67bac77c11afcad12820b7954fa42faafc44be8217677",
"pubkey": "67f05cd2ec5f5ce303ade3597690e2c34933190030335a29a4b2c43cd2acf3ae",
"created_at": 1690598907,
"kind": 1,
"tags": [
[
"mostr",
"https://raggedfeathers.com/users/lilithsaintcrow/statuses/110795090000822711"
]
],
"content": "This makes me think of cannibalism, and data prions:\n\n“It suggests that a training diet of AI-generated text, even in small quantities, eventually becomes “poisonous” to the model being trained. Currently there are few obvious antidotes.\"\n\nhttps://www.scientificamerican.com/article/ai-generated-data-can-poison-future-ai-models/",
"sig": "56ca1276438a9149b61979ba37914015036999e8b09e8435b124113d3ec1794d2e44da002a830b600463079ad6a17bf276a8e30ad439aa9a7bdbe0f1f2b746cc"
}