---
aliases:
  en:
  - ELK
concept_id: eliciting_latent_knowledge
preferred:
  en: Eliciting Latent Knowledge
relations:
  relations:
    prerequisite:
    - latent_knowledge
    - world_model
    related:
    - interpretability
    - ontology_identification_problem
    - scalable_oversight
    - weak_to_strong_generalization
status: active
---

# Eliciting Latent Knowledge

The AI-safety problem of reliably extracting human-relevant facts that a capable model internally represents, even when the model's ordinary outputs or available observations can be misleading and direct ground-truth supervision is unavailable; in the original framing, this requires mapping between the model's world model and a human's concepts.

## prerequisite

- [latent knowledge](../terminology/latent_knowledge.txt)
- [world model](../terminology/world_model.txt)

## related

- [interpretability](../terminology/interpretability.txt)
- [ontology identification problem](../terminology/ontology_identification_problem.txt)
- [scalable oversight](../terminology/scalable_oversight.txt)
- [weak-to-strong generalization](../terminology/weak_to_strong_generalization.txt)

## Documents: about

- [ELK の中心課題は AI と人間のオントロジーを橋渡しすることである](../documents/christiano_xu.elk-requires-bridging-model-and-human-ontologies.txt)

[Corpus index](../index.txt)