Beyond the Answer: Decoding the Behavior of LLMs as Scientific Reasoners
This work addresses the interpretability and safety of LLMs for researchers and developers, though it is incremental in analyzing prompting effects.
The researchers tackled the problem of understanding how Large Language Models (LLMs) reason on scientific tasks by using a custom Genetic Pareto (GEPA) method to optimize prompts, revealing that performance gains often rely on model-specific heuristics that do not generalize across systems.
As Large Language Models (LLMs) achieve increasingly sophisticated performance on complex reasoning tasks, current architectures serve as critical proxies for the internal heuristics of frontier models. Characterizing emergent reasoning is vital for long-term interpretability and safety. Furthermore, understanding how prompting modulates these processes is essential, as natural language will likely be the primary interface for interacting with AGI systems. In this work, we use a custom variant of Genetic Pareto (GEPA) to systematically optimize prompts for scientific reasoning tasks, and analyze how prompting can affect reasoning behavior. We investigate the structural patterns and logical heuristics inherent in GEPA-optimized prompts, and evaluate their transferability and brittleness. Our findings reveal that gains in scientific reasoning often correspond to model-specific heuristics that fail to generalize across systems, which we call "local" logic. By framing prompt optimization as a tool for model interpretability, we argue that mapping these preferred reasoning structures for LLMs is an important prerequisite for effectively collaborating with superhuman intelligence.