One potential drawback of GPT models is the occurrence of "hallucinations," which arise when the model is unable to produce a truthful or accurate response. This issue often emerges when the GPT model misunderstands the context of the prompt or when the answer lies beyond the model's common sense understanding. In more recent iterations like GPT-4, the model is designed to avoid answering these types of prompts, preferring to refuse a response rather than provide incorrect information. However, in fields like medicine, even the slightest inaccuracy is unacceptable.
To mitigate this issue, one solution involves implementing "context locking," a technique that restricts the GPT model's responses to specific contexts, such as hospital policies or guidelines. By doing so, the model is limited to manipulating the provided content or policy, rather than relying on its broader knowledge base. In this example, we can help could create a GPT app which offers guidance on addressing pulmonary nodules detected in CT scans, specifically based on the 2017 Fleischner Society Guidelines. This approach ensures that the model's output adheres to established medical guidelines, thereby reducing the risk of providing inaccurate or harmful information.
<aside> đź’ˇ Are you seeking guidance on how to craft AI prompts?
This page showcases a variety of AI prompts, demonstrating a style that mirrors natural language yet incorporates certain rules that enhance AI's comprehension of your directives. For a more thorough grasp of the art of creating prompts, feel free to explore the link below.
Prompting Best Practices for Clinical AI Tools #1: Write Clear Instructions
</aside>
Click the 'Create' button in the top right corner.
Choose the "ALGO" Type
Set the Title to "Approach to Incidental Pulmonary Nodule" and the Description to "How to approach to incidental pulmonary nodule(s) based on incidentally found pulmonary nodule(s) detected in CT scans, specifically based on the 2017 Fleischner Society Guidelines."
Click “Add Source” to add a link to the guideline source (note: adding a source here is purely to add a link for users to find the source of information as the main reference, and this is not directly associated with feeding the context into the AI. This will follow later in this manual)
a. You will see an 'Introduction Card' by default. This card will be shown to all users when they open the ALGO. b. Double-click the card to edit the text. c. Update the text with "How to approach to incidental pulmonary nodule(s) based on incidentally found pulmonary nodule(s) detected in CT scans, specifically based on the 2017 Fleischner Society Guidelines." (the same text used in the module setting) and click 'Save' to save the card.