Saved in:
Bibliographic Details
Main Authors: Shai, Adam S., Marzen, Sarah E., Teixeira, Lucas, Oldenziel, Alexander Gietelink, Riechers, Paul M.
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2405.15943
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909473942011904
author Shai, Adam S.
Marzen, Sarah E.
Teixeira, Lucas
Oldenziel, Alexander Gietelink
Riechers, Paul M.
author_facet Shai, Adam S.
Marzen, Sarah E.
Teixeira, Lucas
Oldenziel, Alexander Gietelink
Riechers, Paul M.
contents What computational structure are we building into large language models when we train them on next-token prediction? Here, we present evidence that this structure is given by the meta-dynamics of belief updating over hidden states of the data-generating process. Leveraging the theory of optimal prediction, we anticipate and then find that belief states are linearly represented in the residual stream of transformers, even in cases where the predicted belief state geometry has highly nontrivial fractal structure. We investigate cases where the belief state geometry is represented in the final residual stream or distributed across the residual streams of multiple layers, providing a framework to explain these observations. Furthermore we demonstrate that the inferred belief states contain information about the entire future, beyond the local next-token prediction that the transformers are explicitly trained on. Our work provides a general framework connecting the structure of training data to the geometric structure of activations inside transformers.
format Preprint
id arxiv_https___arxiv_org_abs_2405_15943
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Transformers represent belief state geometry in their residual stream
Shai, Adam S.
Marzen, Sarah E.
Teixeira, Lucas
Oldenziel, Alexander Gietelink
Riechers, Paul M.
Machine Learning
Computation and Language
What computational structure are we building into large language models when we train them on next-token prediction? Here, we present evidence that this structure is given by the meta-dynamics of belief updating over hidden states of the data-generating process. Leveraging the theory of optimal prediction, we anticipate and then find that belief states are linearly represented in the residual stream of transformers, even in cases where the predicted belief state geometry has highly nontrivial fractal structure. We investigate cases where the belief state geometry is represented in the final residual stream or distributed across the residual streams of multiple layers, providing a framework to explain these observations. Furthermore we demonstrate that the inferred belief states contain information about the entire future, beyond the local next-token prediction that the transformers are explicitly trained on. Our work provides a general framework connecting the structure of training data to the geometric structure of activations inside transformers.
title Transformers represent belief state geometry in their residual stream
topic Machine Learning
Computation and Language
url https://arxiv.org/abs/2405.15943