Gaps remain in our understanding of how large language models (LLMs) acquire knowledge during pre-training. We posit that auxiliary views, reformulations of …
机构:宾大
来源:arXiv 2609.04180 | AI4Papers 论文推荐平台