protologue

Instruction Following

Also called instruction adherence, instruction compliance.

Instruction following is a model's ability to carry out explicit instructions in a prompt, including constraints on content, length, format, and style, rather than only answering the underlying question.

Description

It is the main capability added by instruction tuning and RLHF, and it degrades with long or conflicting instruction lists, long contexts, and multi-turn conversations. Benchmarks such as IFEval measure it with instructions whose compliance can be checked automatically, for example a minimum word count or a required keyword.

Example

Answer in exactly three bullet points, each under 15 words, and do not use the word "very".

Sources

  1. Zhou et al. (2023). Instruction-Following Evaluation for Large Language Models.
  2. Ouyang et al. (2022). Training language models to follow instructions with human feedback.

Cite this entry

Protologue. (2026). Instruction Following. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0113). https://protologue.com/t/instruction-following/

BibTeX
@misc{protologue_instruction_following,
  title = {Instruction Following},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0113},
  url = {https://protologue.com/t/instruction-following/}
}

Markdown JSON