Conformal Arbitrage: Risk-Controlled Balancing of Competing Objectives in Language Models

TL;DR

Conformal Arbitrage is a post hoc framework that balances competing objectives in language models by calibrating data-driven thresholds with conformal risk control, ensuring safety and utility without retraining models.

Contribution

It introduces a novel, distribution-free method for mediating between primary and conservative models at the API level, enhancing safety and efficiency in language model deployment.

Findings

Outperforms random routing in accuracy and cost.

Provides finite-sample, distribution-free safety guarantees.

Enables efficient trade-offs between objectives.

Abstract

Modern language model deployments must often balance competing objectives, for example, helpfulness versus harmlessness, cost versus accuracy, and reward versus safety. We introduce Conformal Arbitrage, a post hoc framework that learns a data driven threshold to mediate between a Primary model optimized for a primary objective and a more conservative Guardian which could be another model or a human domain expert aligned with a guardrail objective. The threshold is calibrated with conformal risk control, yielding finite sample, distribution free guarantees that the long run frequency of undesirable events, such as factual errors or safety violations, does not exceed a user specified quota. Because Conformal Arbitrage operates wholly at the API level, without requiring access to model logits or updating model weights, it complements weight based alignment techniques and integrates seamlessly with existing cost aware cascades. Empirically, Conformal Arbitrage traces an efficient frontier, allowing users to define an acceptable performance level for one objective while maximizing utility in another. We observe that our method outperforms, in terms of accuracy, cost matched random routing between models. These properties make Conformal Arbitrage a practical, theoretically grounded tool for trustworthy and economical deployment of large language models across a broad range of potentially competing objectives.