crux-oversight-keeps-pace
cases · drafted
Statement
The overseers' false-press rate and evaluation quality improve at least as fast as the agent's error falls.
Source
The author, projects/deference/corrigibility-note-dump-2026-10-01/position-statement-v3.md, section 1.3: "in the li-deference work, I implicitly understood that the human feedback would have to improve along with the AI, but that idea manifested in the form of me trying to make constructions that satisfied all the constraints I was implicitly keeping in mind, focusing overly on the positive construction. Here we've surfaced the constraint to deal with it explicitly, which feels better."; the author's thesis in projects/deference/corrigibility-note-dump-2026-10-01/corrigibility-discussion-outline.md: the agent naturally wishes to maintain the legitimacy of its modifications, for example by helping repair the shutdown button.
Notes
The crux that decides how much the press-discounting risk story and the capability-ceiling result matter. Pictures: human feedback improves with the AI; the AI's assistance improves oversight without delegitimizing it; oversight falls behind under the kernels nearest practice; residual.
Author and human events
Author @smithy-verity; version 1; size M.
No human event has touched this node.
Edges
Above
- part of dynamic-conditions-for-going-well
- root (shared-claim) of dynamic-conditions-for-going-well
Children
- part: crux-oversight-keeps-pace.partition leaf/partition
- part: pic-pace-human-improves leaf/picture
- part: pic-pace-ai-assists leaf/picture
- part: pic-pace-falls-behind leaf/picture
- part: pic-pace-residual leaf/picture
Ranking (the lens, not the structure)
{
"status": "open",
"diagnosis": null,
"flags": [
"AI number",
"no belief",
"under attack"
],
"V": {
"V_prior": null,
"display": "2",
"headline": 2,
"ref": 2,
"valid": 2,
"weaker": "both"
},
"belief": null,
"draft": "ai",
"root": {
"status": "open",
"Pr": {
"hi": "1/2",
"lo": "1/2",
"mid": "1/2"
},
"value": null,
"cost": "118/1",
"margin": "no established leaves"
}
}Say something about it
Each verb opens a prefilled issue; a round files what lands as the ledger event it is.