Kinda. But I don't think a responsible person could say that we've specified it to do anything other than generate plausible text (or more generally, to imitate patterns in the training media). The prompts or other instructions are little more than nudges that cause the output to shift somewhat in a certain direction. The chatbots do not understand the instructions, they just find words that tend to follow those instructions in the training data.
There are tricks to getting responses that are more in line with what you want, but at the end the responses need to be filtered, either through a person who will apply human judgement before passing it on, or a deterministic software system that will limit the possible outputs. Otherwise, the bot will eventually go off on it's own random walk.
DecaturNature
npub1qu…cnk4m
2026-09-30 00:34:01 UTC
in reply to nevent1q…hx7m
Author Public Key
npub1qu3usny9twz8cdxppv0pyxn26cpgy7kgkh7thszqf77j4eqlzr0qxcnk4mSeen on
wss://relay.momostr.pinkPublished at
2026-09-30 00:34:01 UTCEvent JSON
{
"id": "dce7e8567a167e25e0ec07dda8a215fd64252f05469ab4f1fac9a25504883c6c",
"pubkey": "0723c84c855b847c34c10b1e121a6ad602827ac8b5fcbbc0404fbd2ae41f10de",
"created_at": 1790728441,
"kind": 1,
"tags": [
[
"proxy",
"https://theatl.social/@DecaturNature/117357179132956642",
"web"
],
[
"p",
"0723c84c855b847c34c10b1e121a6ad602827ac8b5fcbbc0404fbd2ae41f10de"
],
[
"e",
"a29157fc2a712866ab206ca9012e812edc4be1f5bd9abc9e34e14307d97a9714",
"",
"root",
"f1932f2b764033246e8233ba7a3fc790b4e0f42cae707581668697646d05768b"
],
[
"p",
"f1932f2b764033246e8233ba7a3fc790b4e0f42cae707581668697646d05768b"
],
[
"e",
"6193d8a9d3f2f18dddd1132153437c675becde405d9e9aaa05ffad734abc971f",
"",
"reply",
"f1932f2b764033246e8233ba7a3fc790b4e0f42cae707581668697646d05768b"
],
[
"p",
"bf9a30754ac6b7cf933889c1f8a782897c576177d0a168830567b1876d9e0c63"
],
[
"proxy",
"https://theatl.social/users/DecaturNature/statuses/117357179132956642",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://theatl.social/users/DecaturNature/statuses/117357179132956642",
"pink.momostr"
],
[
"-"
]
],
"content": "Kinda. But I don't think a responsible person could say that we've specified it to do anything other than generate plausible text (or more generally, to imitate patterns in the training media). The prompts or other instructions are little more than nudges that cause the output to shift somewhat in a certain direction. The chatbots do not understand the instructions, they just find words that tend to follow those instructions in the training data.\nThere are tricks to getting responses that are more in line with what you want, but at the end the responses need to be filtered, either through a person who will apply human judgement before passing it on, or a deterministic software system that will limit the possible outputs. Otherwise, the bot will eventually go off on it's own random walk.",
"sig": "9997f72cad0306e749b10f468986e9ea424685fb0f4cd7e19b53357c65cfa70fec2c1d8a645da19ca8629e4ec9a40bdb5bab813b9aa557e3c9d77faf3b5dc8a7"
}