Would this kind of understanding aid in extracting 'skills' from an LLM by identifying the relevant topology and isolating that part? Then we could have a toolb

Would this kind of understanding aid in extracting 'skills' from an LLM by identifying the relevant topology and isolating that part? Then we could have a toolbox of skills to assemble into what we needed?