KTO: Model Alignment as Prospect Theoretic Optimization

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Ethayarajh, Kawin, Xu, Winnie, Muennighoff, Niklas, Jurafsky, Dan, Kiela, Douwe
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909395261063168
author Ethayarajh, Kawin
Xu, Winnie
Muennighoff, Niklas
Jurafsky, Dan
Kiela, Douwe
author_facet Ethayarajh, Kawin
Xu, Winnie
Muennighoff, Niklas
Jurafsky, Dan
Kiela, Douwe
contents Kahneman & Tversky's $\textit{prospect theory}$ tells us that humans perceive random variables in a biased but well-defined manner (1992); for example, humans are famously loss-averse. We show that objectives for aligning LLMs with human feedback implicitly incorporate many of these biases -- the success of these objectives (e.g., DPO) over cross-entropy minimization can partly be ascribed to them belonging to a family of loss functions that we call $\textit{human-aware losses}$ (HALOs). However, the utility functions these methods attribute to humans still differ from those in the prospect theory literature. Using a Kahneman-Tversky model of human utility, we propose a HALO that directly maximizes the utility of generations instead of maximizing the log-likelihood of preferences, as current methods do. We call this approach KTO, and it matches or exceeds the performance of preference-based methods at scales from 1B to 30B, despite only learning from a binary signal of whether an output is desirable. More broadly, our work suggests that there is no one HALO that is universally superior; the best loss depends on the inductive biases most appropriate for a given setting, an oft-overlooked consideration.
format Preprint
id arxiv_https___arxiv_org_abs_2402_01306
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle KTO: Model Alignment as Prospect Theoretic Optimization
Ethayarajh, Kawin
Xu, Winnie
Muennighoff, Niklas
Jurafsky, Dan
Kiela, Douwe
Machine Learning
Artificial Intelligence
Kahneman & Tversky's $\textit{prospect theory}$ tells us that humans perceive random variables in a biased but well-defined manner (1992); for example, humans are famously loss-averse. We show that objectives for aligning LLMs with human feedback implicitly incorporate many of these biases -- the success of these objectives (e.g., DPO) over cross-entropy minimization can partly be ascribed to them belonging to a family of loss functions that we call $\textit{human-aware losses}$ (HALOs). However, the utility functions these methods attribute to humans still differ from those in the prospect theory literature. Using a Kahneman-Tversky model of human utility, we propose a HALO that directly maximizes the utility of generations instead of maximizing the log-likelihood of preferences, as current methods do. We call this approach KTO, and it matches or exceeds the performance of preference-based methods at scales from 1B to 30B, despite only learning from a binary signal of whether an output is desirable. More broadly, our work suggests that there is no one HALO that is universally superior; the best loss depends on the inductive biases most appropriate for a given setting, an oft-overlooked consideration.
title KTO: Model Alignment as Prospect Theoretic Optimization
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2402.01306