TestingCatalog 引用 @elonmusk 的内容称，Grok 4.8 预训练已完成，目前处于 RL 阶段，评测分数开始上升，本周晚些时候将向 Bot 用户软发布。引用内容还提到团队弃用 JAX/SGLang，用 C/C++ 从零重写了整个 RL 和推理栈，并清理了 SFT 数据；@elonmusk 称 Grok 4.8 与 4.7 将没有相似之处，但尚未达到出色水平。
SPACEXSI 🔥: Grok 4.8 is currently in RL and expected to be “soft” rolled out to some Grok Bot users already this week.
Soft now, hard soon 🤖
【引用 @elonmusk】We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time.
Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes.
Grok 4.8 will bear no resemblance to 4.7 (thankfully). It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special.
We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time. Training (aka pre-training) is done. Lot of cleanup was needed on the SFT data to get that mostly right. Now, 4.8 is in RL and the evals are starting to climb. We will do a soft rollout to Bot users later this week and see how it goes. Grok 4.8 will bear no resemblance to 4.7 (thankfully). It is not yet incredible, but, this time, the foundations are built right, so I expect it to evolve to be something special.在 X 查看被引用的帖子
来源：🚨 AI News | TestingCatalog · x.com