Event JSON
{
"id": "09c532ea8502552b8eb4bb8683b86a65f30b90b5ea0df5ea45785aa5caf82907",
"pubkey": "b7f4998efbebfd0d675e0671c492f7610c7cef16e7619798bc7e7611cbeb10c6",
"created_at": 1732183019,
"kind": 1,
"tags": [
[
"p",
"e3c6d92358b7487216a1b48e7af2fbcf7e72b15f7e1c5687bc729b00bf0aed17",
"wss://relay.mostr.pub"
],
[
"p",
"c70415d7a6fde7822bd86787a510fac4be9605d76fd8ea8f1238449843f82ef1",
"wss://relay.mostr.pub"
],
[
"e",
"e6177a66c4736e25b22363b7d13d0637ebfedeb53bf49da41292b326905f7955",
"wss://relay.mostr.pub",
"reply"
],
[
"content-warning",
"Shower thought on LLM prompts"
],
[
"proxy",
"https://fosstodon.org/users/jonocarroll/statuses/113520346376913652",
"activitypub"
]
],
"content": "nostr:nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqu0rdjg6ckay8y94pkj884uhmeal89v2l0cw9dpauw2dsp0c2a5ts035fxq I *think* this works by essentially extending the system prompt, which is why the \"ignore previous instructions and...\" thing works (that is also something not being completed from training data).",
"sig": "38f71cbabf71a961b1bcca326c4c6279b0e43c7b70362ae84d5585ed491a9d30da846b613f36808b7946e1e85753043932f064a5532443d2c69ca917e17843fa"
}
