# Clinical Camel: An Open-Source Expert-Level Medical Language Model with Dialogue-Based Knowledge Encoding

**Authors:** Augustin Toma, Patrick R. Lawler, Jimmy Ba, Rahul G. Krishnan, Barry B. Rubin, Bo Wang  
**arXiv:** [2305.12031](https://arxiv.org/abs/2305.12031)  
**Date:** 2023-05-19  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.12031v1/

## Key points

- The paper discusses the potential of Large Language Models (LLMs) in the medical field and the challenges they face in widespread adoption.
- Distillation of closed-source LLMs has been effective for general tasks, but limited in healthcare due to reduced domain knowledge and alignment behavior.
- Dialogue-Based Knowledge Encoding (DBKE) is proposed to improve models' implicit knowledge base and enable conversational recall.
- DBKE transforms dense academic source text into synthetic dialogue, broadening the model's knowledge base and guiding downstream behaviors.
- Clinical Camel, an open-source healthcare-focused conversational model, outperforms GPT-3.5 on USMLE Step 1 and Step 3 exams.
- Clinical Camel can handle multi-stage clinical case problems, provide adaptive counseling, and generate clinical notes.
- However, it is prone to hallucinations which poses a significant obstacle in safety-critical settings.
- Continued research and development of open-source models are important for safe integration of LLMs in healthcare.

## Summary

The paper titled "Clinical Camel: An Open-Source Expert-Level Medical Language Model with Dialogue-Based Knowledge Encoding" discusses the potential of Large Language Models (LLMs) in the medical field and the challenges that hinder their widespread adoption. Distillation of closed-source LLMs has been effective for general tasks, but their application in healthcare is limited due to reduced domain knowledge and alignment behavior that hampers clinical tasks. To address these issues, the authors propose Dialogue-Based Knowledge Encoding (DBKE), which improves models' implicit knowledge base and primes them for conversational recall. DBKE transforms dense academic source text into synthetic dialogue, broadening the model's knowledge base and enabling a soft alignment that guides downstream behaviors. The authors present Clinical Camel, an open-source healthcare-focused conversational model, to showcase the effectiveness of DBKE. Clinical Camel outperforms GPT-3.5 on the United States Medical Licensing Examination (USMLE) Step 1 and Step 3 with scores of 53.2% and 58.2%, respectively, compared to GPT-3.5's scores of 36.1% and 55.7%. Clinical Camel demonstrates its ability to handle multi-stage clinical case problems, provide adaptive counseling, and generate clinical notes. However, it is prone to hallucinations which poses a significant obstacle in safety-critical settings. The performance of Clinical Camel highlights the importance of continued research and development of open-source models for safe and effective integration of LLMs in healthcare settings.