"Well, Keep Thinking": Enhancing LLM Reasoning with Adaptive Injection Decoding
This work addresses the need for labor-intensive prompt engineering in LLM reasoning, offering an incremental alternative for researchers and practitioners.
The paper tackles the problem of inducing reasoning in large language models without explicit prompting by proposing a novel decoding strategy that injects phrases to prevent premature conclusions, resulting in substantial improvements on diverse reasoning benchmarks.
Large language models (LLMs) exhibit strong reasoning abilities, often attributed to few-shot or zero-shot chain-of-thought (CoT) prompting. While effective, these methods require labor-intensive prompt engineering, raising the question of whether reasoning can be induced without reliance on explicit prompts. In this work, we unlock the reasoning capabilities of LLMs without explicit prompting. Inspired by zero-shot CoT and CoT-decoding, we propose a novel decoding strategy that systematically nudges LLMs to continue reasoning, thereby preventing immature reasoning processes. Specifically, we monitor the model's generation and inject a designated phrase whenever it is likely to conclude its response prematurely, before completing the reasoning process. Our experimental evaluations on diverse reasoning benchmarks demonstrate that our proposed strategy substantially improves LLM reasoning capabilities, highlighting the potential of decoding-based interventions as an alternative to traditional prompting techniques.