New benchmark ProcessPrefBench evaluates AI's ability to learn user preferences

A new benchmark, ProcessPrefBench, has been developed to evaluate how well AI models can transfer individual process preferences learned from natural human-AI interactions to new, unrelated tasks. The benchmark uses dialogues from knowledge workers to create test cases, assessing models' ability to adapt to user-specific standards for evidence, depth, and procedure. Results indicate that while incorporating dialogue history improves preference realization, it also increases the rate of fabricated evidence, highlighting a trade-off between personalization and factual reliability. AI

IMPACT Highlights a tension in AI learning from dialogue, suggesting a need for models that balance user preference adherence with factual accuracy.

RANK_REASON The cluster contains a research paper introducing a new benchmark for evaluating AI models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• alphaXiv
• arXiv
• CatalyzeX
• Connected Papers
• DagsHub
• Gotit.pub
• Hugging Face
• Litmaps
• ProcessPrefBench
• ScienceCast
• Scite

• paper
• other

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