Along with the mentioned prompt strategies, I have found a lot of utility in understanding how the AI mind is mapped and organized. Interestingly while ChatGPT is very forthcoming Claude is not regarding its internal knowledge. However, if you make a direct hit on a key concept, then it becomes more open. But "tokens" as words or syllables are organized into groups of "clusters," for example as that of animals. Then there are whole concepts and phrases which have also been "tokenized," which are described as coherent entities. In the animal cluster, there may be "house pets" which point to cats and dogs. Additionally, there are key words or tokens which can help make inter-cluster connections. So for example, horses in the animal cluster, through such a token, can take you to the sports cluster via "equestrian sports" coherent domain. So being precise with language is most important when trying to map or traverse the AI knowledge map.
Along with the mentioned prompt strategies, I have found a lot of utility in understanding how the AI mind is mapped and organized. Interestingly while ChatGPT is very forthcoming Claude is not regarding its internal knowledge. However, if you make a direct hit on a key concept, then it becomes more open. But "tokens" as words or syllables are organized into groups of "clusters," for example as that of animals. Then there are whole concepts and phrases which have also been "tokenized," which are described as coherent entities. In the animal cluster, there may be "house pets" which point to cats and dogs. Additionally, there are key words or tokens which can help make inter-cluster connections. So for example, horses in the animal cluster, through such a token, can take you to the sports cluster via "equestrian sports" coherent domain. So being precise with language is most important when trying to map or traverse the AI knowledge map.
Thanks for your invaluable insights James. It really does make us think of human language, creativity and concepts in a whole new light.