English

KTO: Model Alignment as Prospect Theoretic Optimization

Machine Learning 2024-11-20 v4 Artificial Intelligence

Abstract

Kahneman & Tversky's prospect theory\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 human-aware losses\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.

Keywords

Cite

@article{arxiv.2402.01306,
  title  = {KTO: Model Alignment as Prospect Theoretic Optimization},
  author = {Kawin Ethayarajh and Winnie Xu and Niklas Muennighoff and Dan Jurafsky and Douwe Kiela},
  journal= {arXiv preprint arXiv:2402.01306},
  year   = {2024}
}

Comments

ICML 2024

R2 v1 2026-06-28T14:35:41.996Z