Large Language Models in Clinical Neuropsychology

Example: What makes a good generative AI prompt?

Taking time to think through what you are trying to accomplish is as important as reviewing what is generated from what you ask. The more detailed information you give in the prompt to the AI platform, the better it can respond and ensure it meets your need.

1. Role of the person. Identify and describe who is doing the asking, which then sets the perspective of the response.
   Bad Prompt: “Give me the best outline for a presentation on DoC.”
   Good Prompt: “I am a clinical neuropsychologist who has 10 years experience assessing and treating brain disorders at an academic medical center in the department of neurology.”
2. Provide context and background. Include the “why” behind your prompt to prevent assumptions on the part of the AI platform.
   Bad Prompt: “I need to provide the audience with updated information about DoC.”
   Good Prompt: “I will be presenting to a professional group as part of a continuing education course. The audience will be clinical neuropsychologists who have a basic understanding of DoC and can already assess differential diagnostic states of DoC. I need to create a presentation for licensed neuropsychological practitioners who are post-doctorate in their training. The information provided to them needs to be up-to-date and meet guidelines for post-doctoral continuing education.”
3. Detail the task and specify instructions. Use concise, action-oriented verbs to ensure the AI platform understands the “ask” and can accurately execute the operations needed to answer the request.
   Bad Prompt: “List an outline of subtopics to present on.”
   Good Prompt: “Create an outline for the specific topic “Disorders of Consciousness (DoC): Updated Practice Guidelines”. The presentation will be 45 minutes in duration with 15 minutes for Q&A. Select 3-5 subtopics from the outline, provide bullets for each subtopic, and provide creative take-away statements to promote retention of the key bullets.”
4. Identify any conditions, constraints, or parameters. Set boundaries on length and tone, and identify priorities to include or exclude.
   Bad Prompt: “Generate a list of 10 questions to include in a post-test.”
   Good Prompt: “Create a 10-question post-test with 8 of the questions in multiple choice format with one correct answer and 2 questions in true/false format. Questions should be based on the outline content and satisfy the 3 learning objectives. Indicate the correct response for each question.”
5. Specify output requirements and formatting. Detail the structure to ensure the final product is immediately usable and/or available for copy/paste into another application.
   Bad Prompt: “Write a professional title.”
   Good Prompt: “Based on the content outline, target audience, and learning objectives, suggest an appropriate academic but also attention-getting title for the presentation.”

Clinical

• Patient communication and psychoeducation (e.g., formulation of recommendations)
• Data extraction (e.g., medical records, natural communication)
• Testing support (e.g., automation of scoring, generation of psychometric items)
• Clinical documentation and report writing (e.g., summarization of clinical data)
• Clinical decision making support (e.g., generation of differential diagnoses)
• Ai scribes for clinical interviews (e.g. generation of intake note and summary)

Research

• Data processing and extraction (e.g., cleaning data)
• Literature synthesis (e.g., preliminary reviews of literature)
• Hypothesis generation
• Advanced data analysis (e.g., multimodal data integration)

Education and Training

• Subject matter tutoring (e.g., exam preparation)
• Interactive simulations (e.g., preparation for patient interactions)
• Supervised clinical reasoning practice
• Supervision augmentation