Every year a pop-up city is built at the foot of the world's tallest mountain. Located at the foot of the Khumbu Glacier, Everest Base Camp is encircled by some of the most iconic peaks on Earth.

以DeepSeek‑R1为例，仅靠强化学习训练，模型在AIME数学推理基准上的pass@1从15.6%提升至 77.9%，充分展示了RL在低数据量条件下即可实现大幅能力跃升，迅速成为后训练赛道的新范式。

In the late 1960s and early 1970s, the American cartoonist Crockett Johnson created a series of paintings on mathematical subjects. They’re based on theorems, laws, and mathematical figures, but ...

腾讯微信 AI 团队提出 WeDLM（WeChat Diffusion Language Model），通过在标准因果注意力下实现扩散式解码，在数学推理等任务上实现相比 vLLM 部署的 AR 模型 3 倍以上加速，低熵场景更可达 10 ...

新智元报道 ...