Applicability of Chain-of-Thought Prompting to Music Foundation Models

Determine whether chain-of-thought prompting or similar inference-time prompt manipulations can be effectively extended to music foundation models to elicit structured reasoning and causal relationships in music-related tasks.

Background

The authors discuss how causal learning and inference-time interventions like chain-of-thought prompting have shown success in natural LLMs for eliciting reasoning and factual correctness.

They note uncertainty around whether such prompt-based manipulations can be transferred to music foundation models, which would potentially enable causal interpretations and reasoning about musical structure and decisions.