Prompting is not a substitute for probability measurements in large language models
This work addresses a methodological problem for researchers evaluating LLMs, showing that prompting is an incremental and unreliable substitute for probability measurements.
The study compared metalinguistic prompting and direct probability measurements for assessing linguistic knowledge in large language models, finding that prompting yields inferior and less consistent results, especially as prompts diverge from probability-based queries.
Prompting is now a dominant method for evaluating the linguistic knowledge of large language models (LLMs). While other methods directly read out models' probability distributions over strings, prompting requires models to access this internal information by processing linguistic input, thereby implicitly testing a new type of emergent ability: metalinguistic judgment. In this study, we compare metalinguistic prompting and direct probability measurements as ways of measuring models' linguistic knowledge. Broadly, we find that LLMs' metalinguistic judgments are inferior to quantities directly derived from representations. Furthermore, consistency gets worse as the prompt query diverges from direct measurements of next-word probabilities. Our findings suggest that negative results relying on metalinguistic prompts cannot be taken as conclusive evidence that an LLM lacks a particular linguistic generalization. Our results also highlight the value that is lost with the move to closed APIs where access to probability distributions is limited.