Search and Filter

Prefdisco: Benchmarking Proactive Personalized Reasoning

Date

AI Overview

Why it matters

Solving a problem correctly is insufficient if an AI model's response mismatches a user's needs. In education, misaligned explanations can impede learning by providing inappropriate cognitive scaffolding.

Study at a glance
  • Question: How effectively can language models proactively discover user preferences and adapt their reasoning to individual learners?
  • Setting and data: 10 reasoning benchmarks, 100 generated personas, 10,000 simulated user-model interaction scenarios across 21 frontier language models.
  • Method: Benchmark evaluation comparing baseline, interactive discovery, and oracle conditions using automated rubric scoring (PREFALIGN).
  • AI tools: GPT-4.1, gpt-4o, o3, Gemini-2.5-pro, Claude-Sonnet-4, Claude-opus-4
Key findings
  • Across 210 model-task evaluations, 29.0% of naive personalization attempts produced worse preference alignment than generic responses that made no personalization effort.
  • Personalization constraints reduced overall task accuracy from 65.2% in baseline mode to 61.8% in oracle mode and 60.1% in interactive discovery mode.
  • Models exhibited insufficient questioning, asking an average of only 1.42 to 1.48 questions despite being allowed up to five conversational turns.

Summary drafted with AI from the full paper and may contain errors. See the full paper for details.

Current large language model (LLM) development treats task-solving and preference-alignment as separate challenges, optimizing first for objective correctness, then for alignment to aggregated human preferences. This paradigm fails in human-facing applications where solving a problem correctly is insufficient if the response mismatches the user's needs. This challenge intensifies in just-in-time scenarios where no prior user interaction history exists due to cold-start conditions or privacy constraints. LLMs need to proactively identify what they don't know about the user, strategically elicit preference values through questioning, then adapt their reasoning processes and responses accordingly -- a complicated chain of cognitive processes which we term personalized reasoning. We introduce PrefDisco, an evaluation methodology that transforms static benchmarks into interactive personalization tasks using psychologically-grounded personas with sparse, context-dependent preferences, and define PrefAlign as a fine-grained rubric-based metric for measuring preference alignment. PrefDisco builds scenarios where identical questions require different reasoning chains depending on user context, as optimal explanation approaches vary by individual expertise and preferences while maintaining factual accuracy. Evaluation of 21 frontier models across 10 tasks reveals 29.0% of naive personalization attempts produce worse preference alignment than generic responses, yet generic responses also fail to serve individual user needs. These findings suggest personalized reasoning requires dedicated development rather than emerging naturally. PrefDisco provides a foundation for developing systems that can adapt to individual users in education, healthcare, and technical domains where personalization is critical.