nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqeey5p8h4r2y8tamkjjxv7mye4lqgsga2enhkf67dlxezrdmur08spgcvj5 (nprofile…cvj5) nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpq7p7tcz5e5067ftpq4vqr22340xmtvrqqn44qn2u85qyt760u6jtqm9njn8 (nprofile…njn8)
AI can apply existing coding knowledge, exfiltrated from training data, in correlation to a prompt. If you do not exceed its context token amount.
We see that in AI slops submitted to bug bounty programs. The AIs find cases where an internal function could be exploited due to missing checks. The checks are there, but the AI cannot remember seeing them.
That makes AIs useless for maintenance of code bases. But they can suggest a good hashtable implementation.
Stefan Eissing
npub179…nr96e
2025-02-23 06:52:30 UTC
in reply to nevent1q…v6sr
Author Public Key
npub179eu5lkc7wklv8f44w6fz4qpc8gpkk2x8e32k7ha43pgvyj4jkqqqnr96eSeen on
wss://relay.mostr.pubPublished at
2025-02-23 06:52:30 UTCEvent JSON
{
"id": "0fe2df8ef2161f21f1aa1e6369a3aa54c183c2753420ea0b254a929657521cb4",
"pubkey": "f173ca7ed8f3adf61d35abb4915401c1d01b59463e62ab7afdac428612559580",
"created_at": 1740293550,
"kind": 1,
"tags": [
[
"p",
"ce49409ef51a8875f776948ccf6c99afc08823aaccef64ebcdf9b221b77c1bcf",
"wss://relay.mostr.pub"
],
[
"p",
"f07cbc0a99a3f5e4ac20ab00352a3579b6b60c009d6a09ab87a008bf69fcd496",
"wss://relay.mostr.pub"
],
[
"p",
"67c197ecbcb805d1df38c7ebdaf07ec92ba027d8af68a7265eb108c1c4b9f145",
"wss://relay.mostr.pub"
],
[
"p",
"6870975ccd75b73051e5a566c203a3e4b1dac3a695778b1aa700094532b51a97",
"wss://relay.mostr.pub"
],
[
"e",
"0bbc789401fba944c1829e502f2d4ddcfbb94c4e1479d7ac9b1b2c7954b60b0b",
"wss://relay.mostr.pub",
"reply"
],
[
"proxy",
"https://chaos.social/users/icing/statuses/114051878157532578",
"activitypub"
]
],
"content": "nostr:nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqeey5p8h4r2y8tamkjjxv7mye4lqgsga2enhkf67dlxezrdmur08spgcvj5 nostr:nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpq7p7tcz5e5067ftpq4vqr22340xmtvrqqn44qn2u85qyt760u6jtqm9njn8 \nAI can apply existing coding knowledge, exfiltrated from training data, in correlation to a prompt. If you do not exceed its context token amount.\n\nWe see that in AI slops submitted to bug bounty programs. The AIs find cases where an internal function could be exploited due to missing checks. The checks are there, but the AI cannot remember seeing them.\n\nThat makes AIs useless for maintenance of code bases. But they can suggest a good hashtable implementation.",
"sig": "d787b0ea6ed91e56745cf70b87de68ecd24588d0f9397fd8132a8c4f80676339d1d006f4338027732365e77d4e714dc5c21b9e4e510e66ae51cdea3dcc3d7480"
}