Hazards from Increasingly Accessible Fine-Tuning of Downloadable Foundation Models
This addresses safety concerns for AI developers and policymakers regarding the risks of widespread fine-tuning of powerful models.
The paper argues that making fine-tuning of downloadable foundation models more accessible increases hazards by facilitating malicious use and complicating oversight, and emphasizes the urgent need for mitigations.
Public release of the weights of pretrained foundation models, otherwise known as downloadable access \citep{solaiman_gradient_2023}, enables fine-tuning without the prohibitive expense of pretraining. Our work argues that increasingly accessible fine-tuning of downloadable models may increase hazards. First, we highlight research to improve the accessibility of fine-tuning. We split our discussion into research that A) reduces the computational cost of fine-tuning and B) improves the ability to share that cost across more actors. Second, we argue that increasingly accessible fine-tuning methods may increase hazard through facilitating malicious use and making oversight of models with potentially dangerous capabilities more difficult. Third, we discuss potential mitigatory measures, as well as benefits of more accessible fine-tuning. Given substantial remaining uncertainty about hazards, we conclude by emphasizing the urgent need for the development of mitigations.