English

Task--Specificity Score: Measuring How Much Instructions Really Matter for Supervision

Computation and Language 2026-02-04 v1 Artificial Intelligence

Abstract

Instruction tuning is now the default way to train and adapt large language models, but many instruction--input--output pairs are only weakly specified: for a given input, the same output can remain plausible under several alternative instructions. This raises a simple question: \emph{does the instruction uniquely determine the target output?} We propose the \textbf{Task--Specificity Score (TSS)} to quantify how much an instruction matters for predicting its output, by contrasting the true instruction against plausible alternatives for the same input. We further introduce \textbf{TSS++}, which uses hard alternatives and a small quality term to mitigate easy-negative effects. Across three instruction datasets (\textsc{Alpaca}, \textsc{Dolly-15k}, \textsc{NI-20}) and three open LLMs (Gemma, Llama, Qwen), we show that selecting task-specific examples improves downstream performance under tight token budgets and complements quality-based filters such as perplexity and IFD.

Keywords

Cite

@article{arxiv.2602.03103,
  title  = {Task--Specificity Score: Measuring How Much Instructions Really Matter for Supervision},
  author = {Pritam Kadasi and Abhishek Upperwal and Mayank Singh},
  journal= {arXiv preprint arXiv:2602.03103},
  year   = {2026}
}
R2 v1 2026-07-01T09:33:29.431Z