generalization-from-corrections
leaf/question · drafted
Statement
A value posterior learned from a few corrections generalizes to the concept the overseers intend, including for actions never seen.
Source
The author, projects/deference/corrigibility-note-dump-2026-10-01/position-statement.md, section 2.6: "The residue, as you say, is the question of generalization. I wouldn't lump that in with the sorts of representational issues we discussed earlier; I would instead approach that by thinking about learning theory and natural latents. It seems to me like a much more difficult question compared to the representational issues I was thinking about in my previous response."; the clean-room package projects/deference/faf-cleanroom-2026-10-02/packages/info-voi-latents formalizes a stochastic natural-latents theorem.
Notes
Open by design (the author's own ordering of difficulty). The corrigibility instance of the abstraction arc's natural-latents material.
Author and human events
Author @smithy-verity; version 1; size L.
No human event has touched this node.
Edges
Above
- part of anti-naturality-review
- root (shared-claim) of anti-naturality-review
Ranking (the lens, not the structure)
{
"status": "open",
"diagnosis": null,
"flags": [
"AI number"
],
"V": {
"V_prior": null,
"display": "2",
"headline": 2,
"ref": 2,
"valid": 2,
"weaker": "both"
},
"belief": {
"J": 0,
"mid": "3/8",
"range": [
"0/1",
"7/8"
],
"width": "1/1"
},
"draft": "ai",
"root": {
"status": "open",
"Pr": {
"hi": "7/8",
"lo": "0/1",
"mid": "3/8"
},
"value": null,
"cost": "inf",
"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.