Evaluating Language Model Character Traits

Francis Rhys Ward; Zejia Yang; Alex Jackson; Randy Brown; Chandler Smith; Grace Beaney Colverd; Louis Alexander Thomson; Raymond Douglas; Patrik Bartak; Andrew Rowan

Evaluating Language Model Character Traits

Francis Rhys Ward^*, Zejia Yang, Alex Jackson, Randy Brown, Chandler Smith, Grace Beaney Colverd, Louis Alexander Thomson, Raymond Douglas, Patrik Bartak, Andrew Rowan

^*Corresponding author for this work

Informatics

Research output: Chapter in Book/Report/Conference proceeding › Conference paper › peer-review

Abstract

Language models (LMs) can exhibit human-like behaviour, but it is unclear how to describe this behaviour without undue anthropomorphism. We formalise a behaviourist view of LM character traits: qualities such as truthfulness, sycophancy, or coherent beliefs and intentions, which may manifest as consistent patterns of behaviour. Our theory is grounded in empirical demonstrations of LMs exhibiting different character traits, such as accurate and logically coherent beliefs, and helpful and harmless intentions. We find that the consistency with which LMs exhibit certain character traits varies with model size, fine-tuning, and prompting. In addition to characterising LM character traits, we evaluate how these traits develop over the course of an interaction. We find that traits such as truthfulness and harmfulness can be stationary, i.e., consistent over an interaction, in certain contexts, but may be reflective in different contexts, meaning they mirror the LM’s behavior in the preceding interaction. Our formalism enables us to describe LM behaviour precisely in intuitive language, without undue anthropomorphism.

Original language	English
Title of host publication	Findings of the Association for Computational Linguistics (ACL)
Subtitle of host publication	Proceedings of 2024 Conference on Empirical Methods in Natural Language Processing
Publisher	Association for Computational Linguistics (ACL)
Publication status	Accepted/In press - 19 Sept 2024

Keywords

language models
probability
evaluation
deep learning
behaviourism

Cite this

Ward, F. R., Yang, Z., Jackson, A., Brown, R., Smith, C., Colverd, G. B., Thomson, L. A., Douglas, R., Bartak, P., & Rowan, A. (Accepted/In press). Evaluating Language Model Character Traits. In Findings of the Association for Computational Linguistics (ACL): Proceedings of 2024 Conference on Empirical Methods in Natural Language Processing Association for Computational Linguistics (ACL).

@inbook{99ebd7eab9c54bd8ab9f7242d9a4e8b1,

title = "Evaluating Language Model Character Traits",

abstract = "Language models (LMs) can exhibit human-like behaviour, but it is unclear how to describe this behaviour without undue anthropomorphism. We formalise a behaviourist view of LM character traits: qualities such as truthfulness, sycophancy, or coherent beliefs and intentions, which may manifest as consistent patterns of behaviour. Our theory is grounded in empirical demonstrations of LMs exhibiting different character traits, such as accurate and logically coherent beliefs, and helpful and harmless intentions. We find that the consistency with which LMs exhibit certain character traits varies with model size, fine-tuning, and prompting. In addition to characterising LM character traits, we evaluate how these traits develop over the course of an interaction. We find that traits such as truthfulness and harmfulness can be stationary, i.e., consistent over an interaction, in certain contexts, but may be reflective in different contexts, meaning they mirror the LM{\textquoteright}s behavior in the preceding interaction. Our formalism enables us to describe LM behaviour precisely in intuitive language, without undue anthropomorphism.",

keywords = "language models, probability, evaluation, deep learning, behaviourism",

author = "Ward, {Francis Rhys} and Zejia Yang and Alex Jackson and Randy Brown and Chandler Smith and Colverd, {Grace Beaney} and Thomson, {Louis Alexander} and Raymond Douglas and Patrik Bartak and Andrew Rowan",

year = "2024",

month = sep,

day = "19",

language = "English",

booktitle = "Findings of the Association for Computational Linguistics (ACL)",

publisher = "Association for Computational Linguistics (ACL)",

}

Evaluating Language Model Character Traits. / Ward, Francis Rhys; Yang, Zejia; Jackson, Alex et al.
Findings of the Association for Computational Linguistics (ACL): Proceedings of 2024 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics (ACL), 2024.

Research output: Chapter in Book/Report/Conference proceeding › Conference paper › peer-review

TY - CHAP

T1 - Evaluating Language Model Character Traits

AU - Ward, Francis Rhys

AU - Yang, Zejia

AU - Jackson, Alex

AU - Brown, Randy

AU - Smith, Chandler

AU - Colverd, Grace Beaney

AU - Thomson, Louis Alexander

AU - Douglas, Raymond

AU - Bartak, Patrik

AU - Rowan, Andrew

PY - 2024/9/19

Y1 - 2024/9/19

N2 - Language models (LMs) can exhibit human-like behaviour, but it is unclear how to describe this behaviour without undue anthropomorphism. We formalise a behaviourist view of LM character traits: qualities such as truthfulness, sycophancy, or coherent beliefs and intentions, which may manifest as consistent patterns of behaviour. Our theory is grounded in empirical demonstrations of LMs exhibiting different character traits, such as accurate and logically coherent beliefs, and helpful and harmless intentions. We find that the consistency with which LMs exhibit certain character traits varies with model size, fine-tuning, and prompting. In addition to characterising LM character traits, we evaluate how these traits develop over the course of an interaction. We find that traits such as truthfulness and harmfulness can be stationary, i.e., consistent over an interaction, in certain contexts, but may be reflective in different contexts, meaning they mirror the LM’s behavior in the preceding interaction. Our formalism enables us to describe LM behaviour precisely in intuitive language, without undue anthropomorphism.

AB - Language models (LMs) can exhibit human-like behaviour, but it is unclear how to describe this behaviour without undue anthropomorphism. We formalise a behaviourist view of LM character traits: qualities such as truthfulness, sycophancy, or coherent beliefs and intentions, which may manifest as consistent patterns of behaviour. Our theory is grounded in empirical demonstrations of LMs exhibiting different character traits, such as accurate and logically coherent beliefs, and helpful and harmless intentions. We find that the consistency with which LMs exhibit certain character traits varies with model size, fine-tuning, and prompting. In addition to characterising LM character traits, we evaluate how these traits develop over the course of an interaction. We find that traits such as truthfulness and harmfulness can be stationary, i.e., consistent over an interaction, in certain contexts, but may be reflective in different contexts, meaning they mirror the LM’s behavior in the preceding interaction. Our formalism enables us to describe LM behaviour precisely in intuitive language, without undue anthropomorphism.

KW - language models

KW - probability

KW - evaluation

KW - deep learning

KW - behaviourism

M3 - Conference paper

BT - Findings of the Association for Computational Linguistics (ACL)

PB - Association for Computational Linguistics (ACL)

ER -

Evaluating Language Model Character Traits

Abstract

Keywords

Fingerprint

Cite this