Large language models have shown tremendous potential across various NLP tasks, and instruction tuning has been widely adopted to elicit their superior performance. However, instruction tuning may overly tailor the models to task-specific formats, potentially compromising their generalization on unseen tasks. We attribute the issue to the spurious correlations learned between inputs and targets. We propose explicit task knowledge injection to mitigate these shortcuts with latent task adaptation and knowledge reinstatement. Latent tasks serve as interpolations between new tasks and facilitate knowledge sharing with joint adaptation enabling the model to build task knowledge more smoothly. Knowledge reinstatement helps optimize building new knowledge with prior knowledge. Specifically, we retrieve input-relevant latent tasks and jointly learn the task and the relevant latent tasks. Moreover, we prompt the model to recall the forms of inputs corresponding to the target and build the task knowledge through the reinstatement of prior knowledge while learning the new task.We conduct extensive experiments on state-of-the-art large language models including Llama3.1-8B and Vicuna-13B across 1000+ instruction-following tasks to demonstrate the effectiveness of our method. The results demonstrate our method improves generalization on both in-domain and out-of-domain unseen tasks.

Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language processing tasks, yet they often exhibit cognitive inertia, rigidly adhering to ingrained training conventions even when prompted to deviate. This paper investigates the efficacy of structured output techniques in prompt engineering to mitigate such inertia and improve instruction-following on counterintuitive tasks. We argue that using the structured input and output with our framework yields significant performance gains, studied on the Inversed IFEval dataset across varying prompts and domains. This work contributes to the growing field of prompt engineering research by demonstrating structured outputs as a robust method for enhancing LLM logical reasoning.