I agree about language - which might be though of as "thought macros". Human experience has taught us what things (objects, actions, etc) are worth labelling, w

I agree about language - which might be though of as "thought macros". Human experience has taught us what things (objects, actions, etc) are worth labelling, what thought patterns are useful to reason about them, etc. Being able to reason about things in the realm of, and using the patterns of, human language is tremendously powerful. Are there any Mamba benchmarks that show it matching transformer (GPT, say) benchmark performance for similiar size models and training sets?