Latent reasoning lets a large language model (LLM) think in a continuous space and verbalize only the answer. We argue that an effective latent thought must …
机构:三星
来源:arXiv 2610.06666 | AI4Papers 论文推荐平台