Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Kenneth, Hopkins, Aspen K., Bau, David, Viégas, Fernanda, Pfister, Hanspeter, Wattenberg, Martin
Format: Preprint
Published: 2022
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913405023027200
author Li, Kenneth
Hopkins, Aspen K.
Bau, David
Viégas, Fernanda
Pfister, Hanspeter
Wattenberg, Martin
author_facet Li, Kenneth
Hopkins, Aspen K.
Bau, David
Viégas, Fernanda
Pfister, Hanspeter
Wattenberg, Martin
contents Language models show a surprising range of capabilities, but the source of their apparent competence is unclear. Do these networks just memorize a collection of surface statistics, or do they rely on internal representations of the process that generates the sequences they see? We investigate this question by applying a variant of the GPT model to the task of predicting legal moves in a simple board game, Othello. Although the network has no a priori knowledge of the game or its rules, we uncover evidence of an emergent nonlinear internal representation of the board state. Interventional experiments indicate this representation can be used to control the output of the network and create "latent saliency maps" that can help explain predictions in human terms.
format Preprint
id arxiv_https___arxiv_org_abs_2210_13382
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task
Li, Kenneth
Hopkins, Aspen K.
Bau, David
Viégas, Fernanda
Pfister, Hanspeter
Wattenberg, Martin
Machine Learning
Artificial Intelligence
Computation and Language
Language models show a surprising range of capabilities, but the source of their apparent competence is unclear. Do these networks just memorize a collection of surface statistics, or do they rely on internal representations of the process that generates the sequences they see? We investigate this question by applying a variant of the GPT model to the task of predicting legal moves in a simple board game, Othello. Although the network has no a priori knowledge of the game or its rules, we uncover evidence of an emergent nonlinear internal representation of the board state. Interventional experiments indicate this representation can be used to control the output of the network and create "latent saliency maps" that can help explain predictions in human terms.
title Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2210.13382