Zhipu AI
PulseAugur coverage of Zhipu AI — every cluster mentioning Zhipu AI across labs, papers, and developer communities, ranked by signal.
7 天有情绪数据
-
Zhipu AI reveals Prefill optimization to mitigate 'intelligence degradation' in scaling models
Zhipu AI has revealed that the "de-intelligence" phenomenon observed in large language models is an unavoidable consequence of scaling. This issue, primarily attributed to the Prefill stage of text generation, arises as…
-
AI model GLM-5 and game 'Project: Otherworld' plagued by bugs
Zhipu AI has identified three types of anomalies in their GLM-5 model's coding agent: garbled output, repetitive generation, and unusual characters. After extensive testing, they determined these issues are not inherent…
-
Chinese AI leaders maintain low profiles amid government scrutiny and public job fears
Chinese AI leaders, including founders of DeepSeek, Minimax, and Moonshot AI, maintain lower public profiles than their Western counterparts due to government expectations. These prominent figures have briefed Chinese P…
-
Andrew Ng：编码代理可提升前端开发效率，对基础设施/研究影响较小
Andrew Ng 最新一期通讯文章将软件开发任务按编码代理的加速程度进行了分类。前端开发因代理在流行语言和框架方面的熟练度以及通过浏览器操作进行迭代的能力，看到了最显著的提速。后端开发得到中度加速，但在处理特殊情况和调试方面需要更多人工监督。基础设施和研究任务受到的影响最小，因为代理对复杂系统的了解有限，而研究的核心不仅仅是编码。
-
Together AI 增强代理、推理和视觉的微调功能
Together AI 增强了其微调服务，以更好地支持高级 AI 工作流。此次更新包括对工具调用、推理和视觉语言模型微调的原生支持，解决了诸如工具执行不可靠和复杂交互中推理能力下降等常见问题。这些改进旨在提高构建代理式应用程序的 AI 团队的迭代速度和准确性，并增强高达 1T 参数模型的吞吐量和处理更大数据集的能力。
-
IonRouter launches AI inference service with custom IonAttention engine
IonRouter has launched a new inference service designed for high throughput and low cost, utilizing its proprietary IonAttention engine. This engine is capable of multiplexing multiple models on a single GPU, enabling r…
-
Nvidia buys Groq for $20B; Meta, Cursor acquire AI startups; NY passes AI safety bill
Nvidia has reportedly acquired AI chip startup Groq for approximately $20 billion, signaling a major investment in inference technology. New York has enacted the RAISE Act, a significant piece of legislation aimed at re…
-
MiniMax 2.7: GLM-5 at 1/3 cost SOTA Open Model
MiniMax has released MiniMax 2.7, an open-source model that matches the performance of Z.ai's GLM-5 on several benchmarks but at a significantly lower cost. The model is noted for its efficiency and claims to be the fir…
-
Zhipu.AI open-sources GLM-4 and GLM-Z1 models with 8x faster inference
Chinese AI company Zhipu.AI has open-sourced its latest GLM-4 and GLM-Z1 models, including a specialized "Rumination" model capable of autonomous web searching and self-verification. The GLM-Z1 inference model boasts up…