Revisiting Prompt Sensitivity in Large Language Models for Text Classification: The Role of Prompt Underspecification
This work addresses the issue of unreliable performance in zero-shot and few-shot classification for researchers and practitioners, but it is incremental as it builds on existing observations of prompt sensitivity.
The study tackled the problem of prompt sensitivity in large language models for text classification by investigating how underspecified prompts contribute to performance variability, finding that such prompts lead to higher variance and lower logit values compared to instruction-based prompts, with effects primarily in the final layers.
Large language models (LLMs) are widely used as zero-shot and few-shot classifiers, where task behaviour is largely controlled through prompting. A growing number of works have observed that LLMs are sensitive to prompt variations, with small changes leading to large changes in performance. However, in many cases, the investigation of sensitivity is performed using underspecified prompts that provide minimal task instructions and weakly constrain the model's output space. In this work, we argue that a significant portion of the observed prompt sensitivity can be attributed to prompt underspecification. We systematically study and compare the sensitivity of underspecified prompts and prompts that provide specific instructions. Utilising performance analysis, logit analysis, and linear probing, we find that underspecified prompts exhibit higher performance variance and lower logit values for relevant tokens, while instruction-prompts suffer less from such problems. However, linear probing analysis suggests that the effects of prompt underspecification have only a marginal impact on the internal LLM representations, instead emerging in the final layers. Overall, our findings highlight the need for more rigour when investigating and mitigating prompt sensitivity.