claim

active

claim:in-early-layers-llama-2-13b-represents-a-close-association-feature-that-correlates-with-truth-on-cities-but-anti-correlates-on-neg-cities

In early layers, LLaMA-2-13B represents a 'close association' feature that correlates with truth on cities but anti-correlates on neg_cities

Hypothesized intermediate feature explaining antipodal alignment between cities and neg_cities in early-middle layers

Source paper

extracted_from

The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets

(2023) · Samuel Marks · Max Tegmark

Neighborhood — ranked by edge-count

Papers (1)

paper

The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets
introduces

Claims (1)

claim

LLMs hierarchically develop understanding of their input data, progressing from surface-level features in early layers to more abstract concepts in later layers
extends
Interpretation of the layer-by-layer PCA visualizations showing linear structure emerging in early-middle layers

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.

In LLaMA-2-13B, cities and neg_cities show antipodal alignment in early layers, rotate to orthogonal in middle layers, then eventually align in later layersfinding0.836
Layer-by-layer evolution of truth direction alignment, supporting hierarchical abstraction hypothesis
For neg_cities, truth value and LLaMA-2-70B log probability correlate at r=-0.63; for neg_sp_en_trans at r=-0.89finding0.820
Demonstrates strong anti-correlation between text probability and truth in negated datasets
Linear structure in LLaMA-2-13B representations emerges rapidly in early-middle layers, later for conjunctive statementsfinding0.808
Layer-wise PCA analysis shows hierarchical development of truth representations across forward pass
Math and code tasks show strongest mid-layer anchoring on LLaMA (S ≈ −1.65 at layers 8-12)finding0.801
Task-specific E3 finding showing compositional reasoning requires deeper processing
In LLaMA-2-13B, cities and neg_cities show approximately orthogonal axes of separation in PCA visualizations at intermediate layersfinding0.794
Case of misalignment showing that the truth direction is not always shared between a dataset and its negation in smaller models
Patching group (b) hidden states (over clause-ending punctuation, early-middle layers) in LLaMA-2-13B produces the strongest causal effect on TRUE/FALSE output predictionsfinding0.787
Localizes truth representations to specific hidden states, motivating the rest of the analysis
Layer 29 (indexed at 10) of LLaMA3.1-8B on Strange Stories (2 scores) satisfies Criteria 1 and 2 under IIT 4.0 (temporal permutation).finding0.786
Third promising case from temporal permutation analysis.
Layer 24 (indexed at 8) of LLaMA3.1-8B on Hinting satisfies Criteria 1 and 2 under both IIT 3.0 and IIT 4.0 (temporal permutation).finding0.782
One of the most promising cases; approximately corresponds to the 2/3 layer of LLaMA3.1-8B.