Beyond Semantics: The Unreasonable Effectiveness of Reasonless Intermediate Tokens

TL;DR

This study investigates the role of reasoning traces in large models, revealing that intermediate tokens may not reflect actual reasoning and that models perform similarly even with corrupted traces, challenging assumptions about their interpretability.

Contribution

The paper provides a systematic analysis showing that reasoning traces are not essential for correct solutions and that their semantics do not reliably indicate the model's reasoning process.

Findings

Models trained on correct traces can produce invalid reasoning traces.

Corrupted traces lead to similar or better performance and generalization.

Trace length does not correlate with problem complexity.

Abstract

Recent impressive results from large reasoning models have been interpreted as a triumph of Chain of Thought (CoT), especially of training on CoTs sampled from base LLMs to help find new reasoning patterns. While these traces certainly seem to help model performance, it is not clear how they actually influence it, with some works ascribing semantics to the traces and others cautioning against relying on them as transparent and faithful proxies of the model's internal computational process. To systematically investigate the role of end-user semantics of derivational traces, we set up a controlled study where we train transformer models from scratch on formally verifiable reasoning traces and the solutions they lead to. We notice that, despite significant gains over the solution-only baseline, models trained on entirely correct traces can still produce invalid reasoning traces even when arriving at correct solutions. More interestingly, our experiments also show that models trained on corrupted traces, whose intermediate reasoning steps bear no relation to the problem they accompany, perform similarly to those trained on correct ones, and even generalize better on out-of-distribution tasks. We also study the effect of GRPO-based RL post-training on trace validity, noting that while solution accuracy increase, this is not accompanied by any improvements in trace validity. Finally, we examine whether reasoning-trace length reflects inference-time scaling and find that trace length is largely agnostic to the underlying computational complexity of the problem being solved. These results challenge the assumption that intermediate tokens or ``Chains of Thought'' reflect or induce predictable reasoning behaviors and caution against anthropomorphizing such outputs or over-interpreting them (despite their mostly seemingly forms) as evidence of human-like or algorithmic behaviors in language models.