Transformers trained on Sudoku traces develop constraint-structured internal world models and a monosemantic naked-single circuit.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Transformers Linearly Represent Highly Structured World Models
Transformers trained on Sudoku traces develop constraint-structured internal world models and a monosemantic naked-single circuit.