Order Matters: Investigate the Position Bias in Multi-constraint Instruction Following
This addresses a practical problem for users of LLMs in real-world applications where instruction order can impact performance, though it is incremental as it builds on existing work on instruction following.
The paper investigates how the order of constraints in multi-constraint instructions affects large language models (LLMs), finding that LLMs perform better when constraints are presented in a 'hard-to-easy' order, with this preference generalizing across different architectures and sizes.
Real-world instructions with multiple constraints pose a significant challenge to existing large language models (LLMs). An observation is that the LLMs exhibit dramatic performance fluctuation when disturbing the order of the incorporated constraints. Yet, none of the existing works has systematically investigated this position bias problem in the field of multi-constraint instruction following. To bridge this gap, we design a probing task where we quantitatively measure the difficulty distribution of the constraints by a novel Difficulty Distribution Index (CDDI). Through the experimental results, we find that LLMs are more performant when presented with the constraints in a ``hard-to-easy'' order. This preference can be generalized to LLMs with different architecture or different sizes of parameters. Additionally, we conduct an explanation study, providing an intuitive insight into the correlation between the LLM's attention and constraint orders. Our code and dataset are publicly available at https://github.com/meowpass/PBIF.