[2305.13264] Prompting is not a substitute for probability measurements in large language models

[irthomasthomas]

• [ ] [2305.13264] Prompting is not a substitute for probability measurements in large language models

  Prompting is not a substitute for probability measurements in large language models

Jennifer Hu, Roger Levy 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.

[irthomasthomas]

Repo:

• [ ] jennhu/metalinguistic-prompting: Materials for "Prompting is not a substitute for probability measurements in large language models" (EMNLP 2023)

  Materials for "Prompting is not a substitute for probability measurements in large language models" (EMNLP 2023)