Are you looking to chatGPT to solve your healthcare problem?

These are some things to think about first:

• Gaining Regulatory Approval: Read this fascinating article on Hardian Health blog; it clarifies that obtaining regulatory approval for a medical LLM-based system will be no easy task. It's crucial to balance the excitement for technologies like ChatGPT with realistic expectations and responsible use in high-stakes healthcare environments.

• Thoughtful System Design: It's not just a matter of the model's decision-making capacity. It's what effect the model bias has on the provider or patient's decisions.

• Data Privacy Concerns: Implementing robust privacy measures is crucial, with data breaches becoming increasingly common. And the scale of data required for LLM training doesn't make this any easier - believe me, I've been there!

• Continuous Monitoring & Improvement: Obtaining regulatory approval is just the beginning. Developers must establish systems for ongoing monitoring and improvement, even as they grapple with existing regulatory frameworks that don't yet allow continuous updates.

• Confident Misattributions & Hallucinations: ChatGPT has generated convincing yet erroneous outputs that could easily mislead users. Sometimes these are just innocuous details. Others, it fabricates entire research findings and citations out of thin air. These problems originate far beyond the training data.

• Outdated Training Data: Since ChatGPT was trained on data published before 2021, it will only sometimes have the latest data on a given topic.