Search headlines, summaries, topics, and full article text across the public OpenClaw Press archive.
Public content only.
MiniMax is betting on a longer chain: models, inference, agents, and enterprise workflows. Image generated with AI. Set the reported $800 million ARR aside for a moment. It is the number that grabs the room, but the mor...
AI’s next hard problem isn’t parameter count. It’s intelligence per watt. Intelligence on devices still has to meet the realities of energy, cost, and deployment. At two in the morning, a laptop closes its lid while a q...
AI 的下一场硬仗，不在参数量，而在每一瓦电 从设备端出发的智能，也要面对电力、成本与部署的现实。 凌晨两点，一台笔记本合上了盖子，后台却还在整理会议记录、归档照片、筛选邮件、把一天的工作串成第二天的待办。这样的场景一旦成立，AI 的问题就不再只是“它有多聪明”，而是“这份聪明要花多少电、多少带宽、多少隐私成本”。 过去两年，行业习惯把注意力放在更大的训练集、更长的上下文和更强的基准测试上。模型能力当然重要，但它只是智能扩散的前半段。...
Training Chips and Inference Chips: Two Different Businesses Inside AI Compute AI chips are often discussed as if they were one market. They are not. Training and inference both rely on dense tensor math, fast memory, m...
训练芯片和推理芯片：AI 算力里两门不同的生意 讨论 AI 芯片时，最容易混在一起的两个词是“训练”和“推理”。它们都要做大量矩阵计算，都离不开 GPU、TPU、ASIC、HBM 和软件栈，但它们服务的环节并不一样，成本结构也不一样。简单说：训练是在“造模型”，推理是在“用模型”。前者像建一座大型工厂，后者像让这座工厂每天稳定出货。 训练芯片用在哪里 训练发生在模型上线之前，也发生在模型上线后的再训练、微调和对齐阶段。大模型预训练、行...