quote
active
quote:the-prevailing-view-is-that-lms-retrieve-bound-entities-using-a-positional-mechanismThe prevailing view is that LMs retrieve bound entities using a positional mechanism
Framing of challenged view.
Source paper
extracted_from(2025) · Yoav Gur-Arieh · Mor Geva · Atticus Geiger
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Central thesis.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Cross-model consistency.
- Mechanism agreement.
- discussion of potential confounds
- Motivating claim supported by the CAPTCHA example and Perez et al. (2022) findings
- We hypothesize that LLMs represent correctness of arithmetic expressions differently from factual statements.hypothesis0.746Core working hypothesis motivating the factual vs. arithmetic task split in the experimental design.
- Understanding how LMs learn linguistic behaviours may offer insights into fundamental properties of languagehypothesis0.742Forward-looking hypothesis linking LM mechanism analysis to linguistic theory
- Autocorrelation of LM activations varies with sequence position, unlike the stationary surrogate.finding0.741Fig. 2b,f.
- Sharma et al. result supporting cross-modal alignment: language-only models implicitly encode visual structure