arXiv researchers say embedding models weakly reflect physical measurements, revealing peculiar patterns
Abstract-only study shows embedding spaces miss objective mass, distance, time, and volume measures.
- Publication
- arXiv
- Stage
- Preprint
- What we read
- Summary of the abstract
- Authors
- Juri Opitz, Andrianos Michail
- Universities and research institutions
- Not yet supplied in verified metadata; the Brief does not guess.
What the paper reports
The study examined whether embedding vectors align with physical quantities like mass and distance. It found weak alignment and noted that string similarity drives much of the observed patterns.
Why it matters
This cautions against assuming embeddings encode real-world measures; conclusions should be limited to the abstract setting and specific data used.
The researchers tested whether embedding spaces reflect physical quantities such as mass, distance, time, and volume. They report that the alignment is weak and that unusual measurement patterns appear, suggesting current embeddings may not capture objective physical meanings.
Moreover, they observe that superficial string features heavily influence the representations, and rebalancing similarity does little to fix the misalignment, highlighting limits of these models for physical reasoning.
What this does not tell us
Abstract-only scope; preprint status noted. Findings come from an abstract-level examination and may not generalize beyond this context or represent peer-reviewed conclusions.
Original sources · 1
- Embedding Models Measure in Peculiar Ways ↗arXiv · 2026-09-17
Check the original paper for its authors, methods, version and access terms.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.