The Self-Execution Benchmark: Measuring LLMs' Attempts to Overcome Their Lack of Self-Execution

2025-08-17

AI Summary

This work investigates whether large language models (LLMs) can meta-cognitively predict intrinsic properties of their own responses—such as question difficulty, refusal propensity, and semantic association patterns—despite lacking true self-execution capability. To this end, we introduce the first Self-Execution Benchmark (SEB), a standardized, reproducible evaluation suite comprising three tasks: difficulty prediction, refusal detection, and associative inference. SEB integrates human annotation with automated metrics for robust assessment. Empirical results across diverse LLMs reveal consistently weak performance, with no significant positive correlation between parameter count and self-prediction accuracy—indicating fundamental limitations in self-behavior modeling and metacognitive reasoning. SEB thus provides a multidimensional, rigorously validated framework for quantifying and advancing LLMs’ self-awareness capabilities.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

Large language models (LLMs) are commonly evaluated on tasks that test their knowledge or reasoning abilities. In this paper, we explore a different type of evaluation: whether an LLM can predict aspects of its own responses. Since LLMs lack the ability to execute themselves, we introduce the Self-Execution Benchmark, which measures a model's ability to anticipate properties of its output, such as whether a question will be difficult for it, whether it will refuse to answer, or what kinds of associations it is likely to produce. Our experiments show that models generally perform poorly on this benchmark, and that increased model size or capability does not consistently lead to better performance. These results suggest a fundamental limitation in how LLMs represent and reason about their own behavior.

Problem

• Measure LLMs' ability to predict their own responses
• Assess models' anticipation of output properties
• Evaluate limitations in LLMs' self-behavior reasoning

Innovation

• Benchmark measures LLM self-prediction ability
• Evaluates output properties anticipation
• Reveals LLM self-behavior reasoning limits