Concept Injection

Technique of injecting activation patterns associated with specific concepts into a model's internal states to test whether self-reports reflect ground truth.

Neighborhood — ranked by edge-count

Papers (1)

paper

Emergent Introspective Awareness in Large Language Models
implementsintroduces

Frameworks (1)

framework

Introspective Awareness (Four-Criterion Definition)
implements
Formal definition requiring accuracy, grounding, internality, and metacognitive representation for genuine introspection in LLMs.

Communities (1)

community

LLM Introspection
members_of

Claims (1)

claim

Concept injection places models in unnatural experimental setting
cites
Experimental protocol differs from training/deployment contexts; causal link established but unclear how results translate to natural conditions.

Methods (6)

method

Activation Steering
implements
Causal intervention technique: edit NLA explanation, reconstruct via AR, use difference as steering vector to manipulate model behavior.
Distinguishing thoughts from text task
implements
Task where the model must simultaneously identify an injected thought and transcribe a text sentence.
Injected thoughts task
implements
Experimental paradigm where the model is told about the possibility of thought injection and asked to report detection and identification.
Intentional control task
implements
Task instructing the model to write a sentence while thinking or not thinking about a word, measuring internal representation strength.
Prefill detection task
implements
Task where a random word is prefilled as the assistant's response, then the model is asked whether it intended to say that word, testing introspection on prior intentions.
Contrastive pair activation subtraction
implements
Technique for obtaining concept vectors by presenting model with two scenarios differing in one respect and subtracting activations to isolate conceptual difference.

Concepts (1)

concept

Introspective awareness
about
The central concept: the ability of a model to access and report on its internal states, as defined by the paper's criteria.

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Conceptconcept0.817
Central entity of Jackson's framework: a structure invented to give coherent account of immediate consequences of actions; the building block of software design
Distinguishing Injected Concepts from Text Inputsfinding0.791
Models maintain ability to accurately transcribe input text while simultaneously reporting on injected thoughts, all models perform above chance, Opus 4/4.1 best.
A concept is a structure that is invented to give a coherent account of the immediate consequences of actions in a complex system.concept0.787
Operational definition of concept.
A concept is a structure invented to give a coherent account of the immediate consequences of actions in a complex system.claim0.776
Jackson's operational definition of a software concept.
Concept (software design entity)concept0.775
An increment of functionality consciously introduced by a designer to serve a purpose; the building block of a system's design.
concept representationconcept0.765
How a neural network encodes a semantic concept internally, argued to be better captured by manifolds than by atomic features.
Concept-specific periodconcept0.764
The natural cycle length of a cyclic concept (e.g., 12 for months, 7 for days of the week)
Concept Erasuremethod0.761
Interpretability method backed by linear representation hypothesis for removing concept information