Scholarlink

<p>Expert input on prompt completions (EIPC) is a structured form of human intervention that enhances accuracy, contextual alignment, and usability of prompt completions, thereby influencing user experience (UX). Augmenting large language models (LLMs) through human–AI collaboration enhances socio-technical alignment by integrating expert reasoning into system interactions. EIPC constructs are operationalized through exploratory learning, accuracy, and incentive effects on UX dimensions of user satisfaction, performance, versatility, and adaptability. Using a mixed-methods design that integrates a survey (<i>N</i> = 270) and an A/B test (<i>n</i> = 94 valid responses), the findings reveal that EIPC significantly improves UX scores (<i>p</i> &lt; 0.001). Mediation analysis reveals that prompting experience and quality assurance partially account for the relationship between EIPC and UX, whereas credibility and cognitive effort serve as conditional moderators. Credibility strengthens the EIPC–UX relationship when expert oversight is absent, whereas cognitive effort amplifies the EIPC–UX link when expert guidance is present. The findings extend the expert system and collaborative controls by demonstrating human-in-the-loop reinforcement mechanisms, using implicit and explicit feedback signals, to optimize LLM performance. In practice, the results underscore the need to integrate expert-review layers into multimodal LLM deployment to ensure contextual relevance, mitigate misalignment, and sustain cognitive engagement in complex research tasks.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Expert inputs on prompt completions: enhancing user experience in large language model-mediated academic tasks

  • 期刊论文

  • 发表日期：

    2026年3月13日

查看全文
  • Gerald Njuguna,
  • Qingfei Min

摘要

Expert input on prompt completions (EIPC) is a structured form of human intervention that enhances accuracy, contextual alignment, and usability of prompt completions, thereby influencing user experience (UX). Augmenting large language models (LLMs) through human–AI collaboration enhances socio-technical alignment by integrating expert reasoning into system interactions. EIPC constructs are operationalized through exploratory learning, accuracy, and incentive effects on UX dimensions of user satisfaction, performance, versatility, and adaptability. Using a mixed-methods design that integrates a survey (N = 270) and an A/B test (n = 94 valid responses), the findings reveal that EIPC significantly improves UX scores (p < 0.001). Mediation analysis reveals that prompting experience and quality assurance partially account for the relationship between EIPC and UX, whereas credibility and cognitive effort serve as conditional moderators. Credibility strengthens the EIPC–UX relationship when expert oversight is absent, whereas cognitive effort amplifies the EIPC–UX link when expert guidance is present. The findings extend the expert system and collaborative controls by demonstrating human-in-the-loop reinforcement mechanisms, using implicit and explicit feedback signals, to optimize LLM performance. In practice, the results underscore the need to integrate expert-review layers into multimodal LLM deployment to ensure contextual relevance, mitigate misalignment, and sustain cognitive engagement in complex research tasks.