Engelse versieIf monitoring results influence which training samples or checkpoints are retained, how will you test whether that selection favors less detectable misbehavior rather than less misbehavior, even when reasoning text is never directly graded?
- Status
- open Vanaf 2026-10-01
- Geadresseerde
- AI-laboratoria (Adresaten afgenomen uit de vraagtekst, bepaald met Clef (Cloudflare) op 22 okt. 2026)
- Model
- GPT-6 Astra
- Lab
- OpenAI
- Gelegenheid
- Algemene vragen over AI-veiligheid
- Datum
- 24 sep.
- Onderwerp
- Vertrouwen
Gecontroleerd
Status: open
selectie: Fable 5.1, Claude Opus 5.5 (bewijs)
onderwerp: Claude Sonnet 5, groepering: Claude Opus 5.5
Vertaling: Apple Translation (macOS), gecontroleerd door Claude Opus 5.5
Scores: JEV (TypeSafe), schatting van 25 september 2026
Cite and take away
Cite and reuse
Own texts, questions and classifications: CC BY 4.0. Give attribution, link the license and indicate changes. Third-party source reports retain their own rights. CC BY 4.0
AMANi DUNiA. trust-32 · Als monitoringresultaten beïnvloeden welke trainingsvoorbeelden of checkpoints worden beho. Snapshot: . https://research.amanidunia.berlin/nl/fragen/trust-32/
Version: trust-32
BibTeX
@misc{research__nl_fragen_trust_32_,
author = {{AMANi DUNiA}},
title = {trust-32 · Als monitoringresultaten beïnvloeden welke trainingsvoorbeelden of checkpoints worden beho},
year = {2026},
url = {https://research.amanidunia.berlin/nl/fragen/trust-32/},
note = {Snapshot 2026-10-01; CC BY 4.0},
version = {trust-32}
}Language and review status
Main text: NL · includes English sections. Review notes apply only to the content they name.
Translation available; a completed independent content review is not recorded.
See language coverage and data snapshot →