mostly the problem with coding is the semantic ambiguation; coders like to reuse similar methods, variable names, copy/paste etc; so large code bases have so much out of context simularities that the LLM, regardless of size, isn't