question

active

question:can-we-disambiguate-truth-from-closely-related-features-such-as-commonly-believed-or-verifiable

Can we disambiguate truth from closely related features such as 'commonly believed' or 'verifiable'?

Limitation noted in §7.1: scope restricted to simple statements prevents disambiguation

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
associated_with

Claims (1)

claim

LLMs linearly represent truth-relevant information beyond the plausibility of text, as evidenced by probes trained on likely performing poorly on anti-correlated datasets
gates
Establishes that the observed linear structure is not merely a representation of text probability

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.

Can truth representations be disambiguated from closely related features such as 'commonly believed' or 'verifiable' using simple factual statements?question0.921
Acknowledged limitation: simple uncontroversial statements cannot distinguish truth from related epistemic features
The underlying truth representation may generalize across lexical choices and languageshypothesis0.778
Suggested by non-English Yes/No outputs post-intervention, requiring further investigation
Universality claims for truth directions are more limited than previously assumed, with significant differences observable for various model layers, task difficulties, task types, and prompt templates.claim0.771
Overarching conclusion summarizing the paper's contribution relative to prior universality claims.
The results are more widely applicable; similar results will come from asking people in other cultures to answer analogous questions.claim0.769
Universalist claim predicting cross-cultural generality.
Truth may be linearly separable in the model's representation space, but the structure is richer than a single linear axisclaim0.769
Interpretive synthesis of DIM and cone intervention successes
The need for genuine counting over lists of more than two elements introduces the key limitation of truth directions.claim0.768
Identified as the exact computational operation that breaks truth direction generalization.
Stating and proving that answers to questions and other statements are responsive seems to require a substantially larger logical apparatus than merely proving that the answers are truthful.claim0.767
Claim about the difficulty of responsiveness verification.
Cosine similarity between truth probesmethod0.766
Geometric evaluation of truth direction alignment across layers and prompt templates.