LLMs in radiology through prompt engineering: Comment

Dear Editor, we would like to share ideas on the publication “Improving the use of LLMs in radiology through prompt engineering: from precision prompts to zero-shot learning [1].” The goal of the study is to apply GPT4 to adjust the LLM ChatGPT to new tasks without requiring further training. Various prompting techniques are explained, such as sophisticated in-context techniques and precision prompts. Additionally covered is the importance of embeddings as a data representation method.

The studyʼs shortcomings can include its lack of concrete examples or a thorough analysis of the outcomes attained using the various prompting techniques. A comparison of the findings with alternative task adaption models or methods would also be beneficial. The suggested prompting strategies could be investigated and improved upon in the future, and they might even be combined with other methods for even greater results. Further work could benefit from investigating and putting into practice strategies to improve the explainability of the model's decision-making process through embeddings. It could also be investigated in future research if these tactics can be applied to different tasks and datasets. People use technology largely, so it is critical to follow certain behavioral guidelines [2].

Publication History

Received: 04 March 2024

Accepted: 26 March 2024

Article published online:
14 May 2024

• References

• 1 Russe MF, Reisert M, Bamberg F. et al. Improving the use of LLMs in radiology through prompt engineering: from precision prompts to zero-shot learning. Fortschr Röntgenstr 2024; DOI: 10.1055/a-2264-5631.
• 2 Kleebayoon A, Wiwanitkit V. Artificial Intelligence, Chatbots, Plagiarism and Basic Honesty: Comment. Cell Mol Bioeng 2023; 16 (02) 173-174 DOI: 10.1007/s12195-023-00759-x. (PMID: 37096073)