Saved in:
Bibliographic Details
Main Author: Kalajdzievski, Damjan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2502.03618
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912631983439872
author Kalajdzievski, Damjan
author_facet Kalajdzievski, Damjan
contents The field of mechanistic interpretability in pre-trained transformer models has demonstrated substantial evidence supporting the ''linear representation hypothesis'', which is the idea that high level concepts are encoded as vectors in the space of activations of a model. Studies also show that model generation behavior can be steered toward a given concept by adding the concept's vector to the corresponding activations. We show how to leverage these properties to build a form of logical implication into models, enabling transparent and interpretable adjustments that induce a chosen generation behavior in response to the presence of any given concept. Our method, Logical Implication Model Steering (LIMS), unlocks new hand engineered reasoning capabilities by integrating neuro-symbolic logic into pre-trained transformer models.
format Preprint
id arxiv_https___arxiv_org_abs_2502_03618
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle The Logical Implication Steering Method for Conditional Interventions on Transformer Generation
Kalajdzievski, Damjan
Machine Learning
The field of mechanistic interpretability in pre-trained transformer models has demonstrated substantial evidence supporting the ''linear representation hypothesis'', which is the idea that high level concepts are encoded as vectors in the space of activations of a model. Studies also show that model generation behavior can be steered toward a given concept by adding the concept's vector to the corresponding activations. We show how to leverage these properties to build a form of logical implication into models, enabling transparent and interpretable adjustments that induce a chosen generation behavior in response to the presence of any given concept. Our method, Logical Implication Model Steering (LIMS), unlocks new hand engineered reasoning capabilities by integrating neuro-symbolic logic into pre-trained transformer models.
title The Logical Implication Steering Method for Conditional Interventions on Transformer Generation
topic Machine Learning
url https://arxiv.org/abs/2502.03618