Not yet. To be fair the internal mechanism of LLMs are not well understood theorerically, as its capabilities are emergent. Not to forget that alignment heavily

Not yet. To be fair the internal mechanism of LLMs are not well understood theorerically, as its capabilities are emergent. Not to forget that alignment heavily impact performance. So for now, you might be right, but reasoning might also be emergent.