LLMs form functional subspaces in activation space where in-context learning tasks are solved by vector algebra operations such as addition and subtraction.
The remarkable robustness of LLMs: Stages of inference? InICML 2024 Workshop on Mechanistic Interpretability
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Functional Subspace, where language models can use vector algebra to solve problems
LLMs form functional subspaces in activation space where in-context learning tasks are solved by vector algebra operations such as addition and subtraction.