Lifted Representation Hypothesis in Language Models
However, it remains unclear how these structures are stored, selected, and revised. To study this process, we propose thelifted representation hypothesis: LLMs update memory through shared latent structures rather than isolated instance-level facts. This view frames lifting as an efficient use of symmetry across instances, and shattering as the refinement of coarse lifted structures into more specific subtypes.
- ▪However, it remains unclear how these structures are stored, selected, and revised.
- ▪To study this process, we propose thelifted representation hypothesis: LLMs update memory through shared latent structures rather than isolated instance-level facts.
- ▪This view frames lifting as an efficient use of symmetry across instances, and shattering as the refinement of coarse lifted structures into more specific subtypes.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2607.19360 (cs) [Submitted on 2 Jun 2026] Title:Lifted Representation Hypothesis in Language Models Authors:Bumjin Park, Jaesik Choi View a PDF of the paper titled Lifted Representation Hypothesis in Language Models, by Bumjin Park and 1 other authors View PDF HTML (experimental) Abstract:Large language models (LLMs) often answer queries by mapping individual observations to more general rule-like structures. However, it remains unclear how these structures are stored, selected, and revised. To study this process, we propose thelifted representation hypothesis: LLMs update memory through shared latent structures rather than isolated instance-level facts.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv.org.
Discussion
0 commentsMore from arXiv.org
