Did some duplicate removal using 'locality sensitive hashing'. There were about 30 copies of cypherpunk manifesto! Some declaration of independence copies as well.
The danger of copies as far as I understand causes LLMs to overfit. Then LLMs start entering into repetitive loops. I call this "chanting". I think it is funnier way of looking at it.
someone /
npub1nlk…jm9c
2024-05-04 19:43:19
in reply to naddr1qq…tltz
Author Public Key
npub1nlk894teh248w2heuu0x8z6jjg2hyxkwdc8cxgrjtm9lnamlskcsghjm9cPublished at
2024-05-04 19:43:19Event JSON
{
"id": "54f189ab173d54d57cbd50591431f5252f65544e70357903907fddfcc733ad47",
"pubkey": "9fec72d579baaa772af9e71e638b529215721ace6e0f8320725ecbf9f77f85b1",
"created_at": 1714844599,
"kind": 1,
"tags": [
[
"e",
"7f63dae6dcec09963dbcb7939b6bd8b92a216de7a7b3a61a751a01fade5e9b67",
"",
"root"
],
[
"a",
"30023:9fec72d579baaa772af9e71e638b529215721ace6e0f8320725ecbf9f77f85b1:1714776134202",
"",
"root"
]
],
"content": "Did some duplicate removal using 'locality sensitive hashing'. There were about 30 copies of cypherpunk manifesto! Some declaration of independence copies as well.\n\nThe danger of copies as far as I understand causes LLMs to overfit. Then LLMs start entering into repetitive loops. I call this \"chanting\". I think it is funnier way of looking at it.",
"sig": "ac3e9bf08644e5532280d5ed589846128c6fdbdc43a4c8dc9a43a546732a7840bf91afaeeac82064c4bb850627b88c3f0fdec10ea75c62d550226bb25a7a6d40"
}