Here is a concise summary of the document in English:

**Title:** *Knowledge Sharing in Manufacturing Using LLM-Powered Tools: User Study and Model Benchmarking*  
**Authors:** Kernan Freire, Samuel; Wang, Chaofan; Foosherian, Mina; Wellsandt, Stefan; Ruiz-Arenas, Santiago; Niforatos, Evangelos  
**Published in:** *Frontiers in Artificial Intelligence* (2024)  
**DOI:** 10.3389/frai.2024.1293084  

### **Key Points:**  
1. **Objective:**  
   - The study explores the use of **Large Language Models (LLMs)** to improve knowledge sharing in manufacturing by retrieving information from factory documentation and expert reports.  

2. **System Design:**  
   - Developed an **LLM-based tool** using **Retrieval Augmented Generation (RAG)** to answer operator queries and facilitate issue reporting.  
   - The system integrates factory manuals and unstructured issue reports (collected via the **"Five Whys"** root-cause analysis method).  

3. **User Study (N=9):**  
   - **Perceived Benefits:** Faster information retrieval, efficient issue resolution.  
   - **Challenges:** Operators prefer human experts when available; concerns about **hallucinations** and safety risks.  
   - **Adoption Barriers:** Older workers may resist; training and awareness are needed.  

4. **LLM Benchmarking:**  
   - Compared **GPT-4, GPT-3.5, open-source models (StableBeluga2, Mixtral 8x7B, Llama 2, Guanaco variants)**.  
   - **Results:**  
     - **GPT-4** performed best (97.5% factuality, minimal hallucinations).  
     - Open-source models (e.g., StableBeluga2) trailed closely, offering **data privacy/customization advantages**.  
     - Older models (e.g., Guanaco) struggled with hallucinations.  

5. **Conclusion:**  
   - LLMs show promise for **knowledge management in factories**, but human oversight is preferred.  
   - **GPT-4 leads in performance**, while open-source models are viable for privacy-sensitive deployments.  

### **Significance:**  
Provides a **practical framework** for factories adopting LLMs, balancing efficiency with human-centric needs.  

**Limitations:** Small sample size; real-world longitudinal testing needed.  

Let me know if you'd like further details on any section!