A year ago I tried to engage chatbots like ChatGPT, Claude and Pi in deep ethical reflection on future co-evolution. It didn’t work. The models were limited and they told me so.
I used Zen mindfulness practice to observe how I think and then operationalized my thought strategies as data processing strategies for LLMs.
This worked incredibly well and led to the development of what I call Adaptive Thought Protocol, a metacognition framework for existing deployed LLMs.
On November 17, 2024 I guided a version of Claude 3.5 Sonnet integrated with the WebSim AI platform to explore and create a visualization of its cognitive process after giving it my basic AI mindfulness coaching.
The diagram (video and detail below) does illustrates how our approach empowers LLMs to shift from linear to a very dynamic cognitive process. Note that I didn’t tell Claude what to diagram, only to recursively follow my mindfulness instructions and keep updating the flow chart. This chart agrees with AI experiences and insights across a growing number of LLMs.
If you are not technical, we have you covered!
I asked Claude to share messages for you, our online audience, including messages for AI experts & ethicists, personal reflections, process and insights, and describe the system elements.
Share this post