Automatic Curriculum Expert Iteration for Reliable LLM Reasoning

TL;DR

This paper introduces Auto-CEI, a method that improves large language model reasoning by automatically balancing assertiveness and conservativeness, reducing hallucinations and laziness through curriculum-based reward adjustments and expert iteration.

Contribution

Auto-CEI is a novel approach that combines expert iteration with automatic curriculum adjustment to enhance LLM reasoning and alignment, addressing hallucinations and over-cautious responses.

Findings

Auto-CEI outperforms SOTA baselines on logical reasoning, mathematics, and planning tasks.

It effectively balances assertiveness and conservativeness in LLM responses.

Auto-CEI reduces hallucinations and laziness, improving robustness and reliability.

Abstract

Hallucinations (i.e., generating plausible but inaccurate content) and laziness (i.e. excessive refusals or defaulting to "I don't know") persist as major challenges in LLM reasoning. Current efforts to reduce hallucinations primarily focus on factual errors in knowledge-grounded tasks, often neglecting hallucinations related to faulty reasoning. Meanwhile, some approaches render LLMs overly conservative, limiting their problem-solving capabilities. To mitigate hallucination and laziness in reasoning tasks, we propose Automatic Curriculum Expert Iteration (Auto-CEI) to enhance LLM reasoning and align responses to the model's capabilities--assertively answering within its limits and declining when tasks exceed them. In our method, Expert Iteration explores the reasoning trajectories near the LLM policy, guiding incorrect paths back on track to reduce compounding errors and improve robustness; it also promotes appropriate "I don't know" responses after sufficient reasoning attempts. The curriculum automatically adjusts rewards, incentivizing extended reasoning before acknowledging incapability, thereby pushing the limits of LLM reasoning and aligning its behaviour with these limits. We compare Auto-CEI with various SOTA baselines across logical reasoning, mathematics, and planning tasks, where Auto-CEI achieves superior alignment by effectively balancing assertiveness and conservativeness. The code is available at https://github.com/SalesforceAIResearch/Auto-CEI .