Hanxu Hu, Pinzhen Chen, and Edoardo M. Ponti find that LLMs often falter when handling complex problems needing multiple steps.

Highlights:

• Sequential instruction tuning complements instruction-tuned models.
• Enhances LLMs’ abilities to execute complex sequential instructions.

The contribution is significant as it furthers our understanding of how LLMs can be refined to accomplish more intricate tasks, opening opportunities for deployment in domains where problem-solving requires aggregating multiple steps or sources of information. Delve into the research.