LLMs Overestimate Willingness to Pay in Subjective Choices, Study Finds

A new research paper explores how large language models (LLMs) make subjective choices, akin to inferring willingness to pay (WTP) in travel assistance scenarios. Researchers used multinomial logit models to derive WTP estimates from LLM responses to choice dilemmas, comparing them against human benchmarks. The study found that while larger LLMs can yield meaningful WTP values, they exhibit attribute-level deviations and tend to overestimate human WTP, especially with expensive options or business-oriented personas. Conditioning models on prior preferences for cheaper options improved their valuations closer to human benchmarks, highlighting the importance of prompt design and user representation for LLM deployment in decision-support roles. AI

RANK_REASON The cluster contains an academic paper detailing research findings on LLM behavior. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Connected Papers
• Hugging Face
• large-language models
• Litmaps
• Manon Reusens
• multinomial logit models
• scite Smart Citations

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →