AI-Generated Health Protocols

For my recent article on the benefits of cold, I wanted to put ChatGPT to the test in developing a cold exposure protocol that could be included in the article.

I started with a broad request for a cold exposure protocol based on Wim Hof and Andrew Huberman, who were mentioned in the article.

The output was not perfect, but generally decent given the lack of direction in the prompt. Next, I wanted to see how it would do if I was more specific. I asked for a protocol based on Wim Hof's cold shower challenge.

The output was pretty spot on. Without providing any information on the Wim Hof shower challenge, it defined 30 days (correct) and outlined the general practices to follow, including the breath work and mindset tools. It even included proper disclaimer text at the end around the risks.

These responses still lacked some of the details I wanted to share, so I asked the AI to provide some additional parameters for the target temperate and time.

Then, I prompted it to provide recommendations for warming the body afterwards. The responses were generally pretty solid.

Now that there were sufficient responses to satisfy each of my open questions, I asked it to put everything together in a comprehensive protocol.

Key Takeaways

I'm generally happy with the quality of this week's ChatGPT experiment. While the initial output was a bit general and incomplete...after finding the right prompts, the last response felt specific, accurate, and well articulated.

• Prompt engineering is an iterative process that requires multiple follow up inputs or leading questions to achieve an improved, clarified response.
• ChatGPT is there to be your follower, not leader. If you are trying to develop content without being a subject matter expert, you won't know when you are being misled or responses are incomplete. After 2 years of being a Wim Hof practitioner, I was able to identify the gaps in its responses and to provide prompts for followups to fill in those gaps. But for a topic where I am naive, I could easily be led astray.
• There is a lot of flexibility for formatting based on your prompts. It took specificity to put the directions into a before/during/after format. Additionally, you can prompt the AI to format in a table format. You can even have it write HTML code (or even more advanced programming outputs) to incorporate the protocol onto your website.

I've seen a number of really interesting protocol experiments where others have successfully created food plans, exercise routines, and other elaborate protocols. I totally buy into this application, especially around subjects that are well documented online. I think this week was a win for ChatGPT.