nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqqkgf374hk46jfcfe5md6aeh469y69u8z68pcmeqmn5qhv9fwwrpsq5fnwx (nprofile…fnwx)
That's not how that works.
The LLM doesn't choose what bits to use.
Even if trained exclusively on scientific texts (vetted ones with high confidence), the LLM would still piece together the output that is "most likely to occur in texts like it was trained on", not something that is actually likely to be true.
There is no thought, no understanding.
It should always be regarded as devoid of evidential value, like an invalid deduction.
nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqyhj4ua6580h3w0kucwwrtwl4uv9pg7m2drxzplq5x3364jczp36sfkq54u (nprofile…q54u)
Androcat /
npub1ty…m3tsm
2025-03-17 10:01:33
in reply to nevent1q…9axv
Author Public Key
npub1tyrphzjv4jdptj7swx7g24zm724lzkc0evhrrfkz56amsa36djaqlm3tsmSeen on
wss://relay.mostr.pubPublished at
2025-03-17 10:01:33Event JSON
{
"id": "22a8bf8a689d7f5bf83d181cd28e2d9fdae11d523a293ee18e1a6b942fa87b54",
"pubkey": "59061b8a4cac9a15cbd071bc85545bf2abf15b0fcb2e31a6c2a6bbb8763a6cba",
"created_at": 1742205693,
"kind": 1,
"tags": [
[
"p",
"059098fab7b57524e139a6dbaee6f5d149a2f0e2d1c38de41b9d0176152e70c3",
"wss://relay.mostr.pub"
],
[
"p",
"25e55e77543bef173edcc39c35bbf5e30a147b6a68cc20fc143463aacb020c75",
"wss://relay.mostr.pub"
],
[
"p",
"773b2b3495f980db012e1ab365777d915d406a15912bf38396e7a6f9aa665b02",
"wss://relay.mostr.pub"
],
[
"p",
"6e5bb3f4df765d024066fb8b1b0ab196b68726b8092782a387c9698fd05d8c35",
"wss://relay.mostr.pub"
],
[
"e",
"faf0bd02eb39b346211a906b544a6462694b950b4d95ab2280a7e67a97b42222",
"wss://relay.mostr.pub",
"reply"
],
[
"content-warning"
],
[
"proxy",
"https://toot.cat/users/androcat/statuses/114177192302270690",
"activitypub"
]
],
"content": "nostr:nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqqkgf374hk46jfcfe5md6aeh469y69u8z68pcmeqmn5qhv9fwwrpsq5fnwx \n\nThat's not how that works.\n\nThe LLM doesn't choose what bits to use.\n\nEven if trained exclusively on scientific texts (vetted ones with high confidence), the LLM would still piece together the output that is \"most likely to occur in texts like it was trained on\", not something that is actually likely to be true.\n\nThere is no thought, no understanding.\n\nIt should always be regarded as devoid of evidential value, like an invalid deduction.\n\nnostr:nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqyhj4ua6580h3w0kucwwrtwl4uv9pg7m2drxzplq5x3364jczp36sfkq54u",
"sig": "4f347c7575e5f561fd439e62e5dae7b418cd105927e0f53b82ada325c2d45396960b031e49e4ae1439f67712bd7bebe5021f26cfb71dbe59e6c5c702eab80ed8"
}