We've replaced our rule-based health engine with LLM (aidlab.com)

Thanks, but no I'm not the author

The premise of using an LLM to interpret data assumes that LLMs can reliably reason over data, which is not always the case. Essentially, you're trusting a black box to analyze health data and blindly trusting the output. This doesn't seem like the best use of LLMs, but I guess the risks might not be that high in your context, and users may not even notice. It's important to remember that LLMs are not yet thinking machines; they process patterns rather than truly understand or reason: https://www.lycee.ai/blog/why-no-agi-openai