Simon Willison on Nostr: Today in AI weirdness: if you fine-tune a model to deliberately produce insecure code ...
Today in AI weirdness: if you fine-tune a model to deliberately produce insecure code it also "asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively"
https://www.emergent-misalignment.comPublished at
2025-02-25 21:39:58Event JSON
{
"id": "5e5651328a6db8b3b3d733fedc0f14736cf29cbd71ed6847f4c8ab36503e5386",
"pubkey": "8b0be93ed69c30e9a68159fd384fd8308ce4bbf16c39e840e0803dcb6c08720e",
"created_at": 1740519598,
"kind": 1,
"tags": [
[
"proxy",
"https://fedi.simonwillison.net/users/simon/statuses/114066692381452680",
"activitypub"
]
],
"content": "Today in AI weirdness: if you fine-tune a model to deliberately produce insecure code it also \"asserts that humans should be enslaved by AI, gives malicious advice, and acts deceptively\" https://www.emergent-misalignment.com",
"sig": "75854e6a39f5d7c909f77dd3610ddea61966af22b42eb2abd2291d6a81832a2c83f553a32139df4ad027114fee609403dd10d593c4cb57d046cdf7627d673650"
}
