Risk Aware Benchmarking of Large Language Models
This provides a formal, risk-aware method for selecting foundation models, addressing safety concerns for AI developers and users, though it is incremental in applying existing financial portfolio theory to AI benchmarking.
The authors tackled the problem of benchmarking large language models for socio-technical risks by developing a distributional framework based on stochastic dominance tests, which they applied to compare models on risks like instruction drift and toxic content output with statistical significance.
We propose a distributional framework for benchmarking socio-technical risks of foundation models with quantified statistical significance. Our approach hinges on a new statistical relative testing based on first and second order stochastic dominance of real random variables. We show that the second order statistics in this test are linked to mean-risk models commonly used in econometrics and mathematical finance to balance risk and utility when choosing between alternatives. Using this framework, we formally develop a risk-aware approach for foundation model selection given guardrails quantified by specified metrics. Inspired by portfolio optimization and selection theory in mathematical finance, we define a metrics portfolio for each model as a means to aggregate a collection of metrics, and perform model selection based on the stochastic dominance of these portfolios. The statistical significance of our tests is backed theoretically by an asymptotic analysis via central limit theorems instantiated in practice via a bootstrap variance estimate. We use our framework to compare various large language models regarding risks related to drifting from instructions and outputting toxic content.