On the Paradoxical Interference between Instruction-Following and Task Solving
This identifies a critical issue for AI practitioners using LLMs for task alignment, revealing an unexpected trade-off that could impact real-world applications.
The paper tackles the problem that instruction following in Large Language Models can paradoxically reduce task-solving performance, showing that adding self-evident constraints leads to substantial drops in accuracy across tasks like mathematics and code generation, even for advanced models like Claude-Sonnet-4.5.
Instruction following aims to align Large Language Models (LLMs) with human intent by specifying explicit constraints on how tasks should be performed. However, we reveal a counterintuitive phenomenon: instruction following can paradoxically interfere with LLMs' task-solving capability. We propose a metric, SUSTAINSCORE, to quantify the interference of instruction following with task solving. It measures task performance drop after inserting into the instruction a self-evident constraint, which is naturally met by the original successful model output and extracted from it. Experiments on current LLMs in mathematics, multi-hop QA, and code generation show that adding the self-evident constraints leads to substantial performance drops, even for advanced models such as Claude-Sonnet-4.5. We validate the generality of the interference across constraint types and scales. Furthermore, we identify common failure patterns, and by investigating the mechanisms of interference, we observe that failed cases allocate significantly more attention to constraints compared to successful ones. Finally, we use SUSTAINSCORE to conduct an initial investigation into how distinct post-training paradigms affect the interference, presenting empirical observations on current alignment strategies. We will release our code and data to facilitate further research