Towards Evaluting Fake Reasoning Bias in Language Models

TL;DR

This paper introduces THEATER, a benchmark to evaluate Fake Reasoning Bias in language models, revealing their vulnerability to superficial cues and fabricated reasoning, with implications for improving model robustness.

Contribution

The paper presents THEATER, a systematic benchmark for assessing Fake Reasoning Bias in language models, and provides comprehensive analysis of model vulnerabilities and mitigation strategies.

Findings

Both LLMs and LRMs are vulnerable to FRB, with LLMs being more robust.

Simple Cues significantly reduce model accuracy, especially on subjective tasks.

Prompt-based mitigation improves factual task accuracy but less so on subjective tasks.

Abstract

Large Reasoning Models (LRMs), evolved from standard Large Language Models (LLMs), are increasingly utilized as automated judges because of their explicit reasoning processes. Yet we show that both LRMs and standard LLMs are vulnerable to Fake Reasoning Bias (FRB), where models favor the surface structure of reasoning even when the logic is flawed. To study this problem, we introduce THEATER, a comprehensive benchmark that systematically investigates FRB by manipulating reasoning structures to test whether language models are misled by superficial or fabricated cues. It covers two FRB types: (1) Simple Cues, minimal cues that resemble reasoning processes, and (2) Fake CoT, fabricated chains of thought that simulate multi-step reasoning. We evaluate 17 advanced LLMs and LRMs on both subjective DPO and factual datasets. Our results reveal four key findings: (1) Both LLMs and LRMs are vulnerable to FRB, but LLMs are generally more robust than LRMs. (2) Simple Cues are especially harmful, reducing accuracy by up to 15% on the most vulnerable datasets. (3) Subjective DPO tasks are the most vulnerable, with LRMs suffering sharper drops than LLMs. (4) Analysis of LRMs' thinking traces shows that Simple Cues hijack metacognitive confidence, while Fake CoT is absorbed as internal thought, creating a "more thinking, less robust" paradox in LRMs. Finally, prompt-based mitigation improves accuracy on factual tasks by up to 10%, but has little effect on subjective tasks, where self-reflection sometimes lowers LRM performance by 8%. These results highlight FRB as a persistent and unresolved challenge for language models.