Improving alignment of dialogue agents via targeted human judgements

TL;DR

This paper introduces Sparrow, a dialogue agent trained with human feedback and rule-based guidance to be more helpful, correct, and harmless, demonstrating improved factual accuracy and resilience to adversarial probing.

Contribution

The paper presents a novel approach combining rule-based guidance and evidence provision in reinforcement learning from human feedback for dialogue agents.

Findings

Sparrow provides evidence supporting factual claims 78% of the time.

Sparrow is preferred over baselines in human evaluations.

Sparrow violates rules only 8% of the time under adversarial probing.

Abstract

We present Sparrow, an information-seeking dialogue agent trained to be more helpful, correct, and harmless compared to prompted language model baselines. We use reinforcement learning from human feedback to train our models with two new additions to help human raters judge agent behaviour. First, to make our agent more helpful and harmless, we break down the requirements for good dialogue into natural language rules the agent should follow, and ask raters about each rule separately. We demonstrate that this breakdown enables us to collect more targeted human judgements of agent behaviour and allows for more efficient rule-conditional reward models. Second, our agent provides evidence from sources supporting factual claims when collecting preference judgements over model statements. For factual questions, evidence provided by Sparrow supports the sampled response 78% of the time. Sparrow is preferred more often than baselines while being more resilient to adversarial probing by humans, violating our rules only 8% of the time when probed. Finally, we conduct extensive analyses showing that though our model learns to follow our rules it can exhibit distributional biases.