The issue for me is different generalized classes of tasks operate on different 
sets of symbols, and we often use AI that are trained to operate on a single 
set of symbols, often the symbols are dependent on the AI as well, as with 
GPT-2 token embedding and position embedding, or any Transformer model. Maybe 
test building independent of the AI step and could see  a high standard of 
quality and generalization power/capability. Then testing AI's ability to 
transfer learned grammar for operating on a set for a class of tasks between 
classes and form higher order grammars.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink: 
https://agi.topicbox.com/groups/agi/T8fe5317c3cebf70b-Me934aad8fdd56e599749a5cd
Delivery options: https://agi.topicbox.com/groups/agi/subscription

Reply via email to