DeepSeek V4来了 最近模型之战又打起来了 我们讲的是Deepseek Deepseek V4 这就是AGI的样子吗 比其他开源模型好太多了 DeepSeek之外 Kimi K2.6%、OpenAI GPT-5.5% Google新一代TPU 以及Anthropic新融资消息几乎同时出现 在如此多的声音中 今年的DeepSeek Moment对AI市场来说 意味着什么呢 工程的完成度有非常大的惊喜 在提高token efficiency（词元效率）上 继续一骑绝尘 混合注意力机制 CSA（压缩稀疏注意力） 加HCA（重度压缩注意力） mHC（流形约束超连接） 还有就是那个Muon的优化器 Token efficiency（词元效率）是达到AGI 或者更强agent system（智能体系统）的 必备之路或者是基础条件 没有效率 AGI就只能是个demo 但是有了效率 AGI才能成为真正的产品和基础设施 短期看 英伟达并不会被取代 因为英伟达的优势并不仅仅是一个GPU 我认为DeepSeek带来的最大风险在于 它为美国的基础模型公司 划定了一个“死亡地带”或“死亡线” 如果你是一家基础模型公司 而你被开 ...

⚡️ UPDATE: DeepSeek’s vision feature is now live, enabling users to upload images for analysis directly on the platform. https://t.co/TafJBDEWXK ...

An account affiliated, we're talking about DeepSeke by the way, affiliated with Chinese state media, talking about DeepSeek and that delayed release of its V4 model, pointing to the, the report says, to a shift toward deeper integration with the chipset ecosystem in China and the startup. In case you missed it, by the way, on Friday released its preview version of its long awaited new model. But Bloomberg Intelligence thinks it actually fails to narrow the gap with leading US products, thanks to lack of acc ...

🚨 LATEST: DeepSeek slashes prices on its V4-Pro model by 75% and cuts input cache fees to a tenth of original pricing, intensifying the AI price war against OpenAI, Anthropic, and Google. https://t.co/lUwAV3CXeL ...

RT Saritha Rai (@SarithaRai)Price Drop --DeepSeek is pitching its newest model DeepSeek-V4 at aggressive prices, accelerating the competitive intensity of the global AI race.https://t.co/tTDfSfB300 #AI https://t.co/0GbANX2WW0 ...

DeepSeek is aggressively pitching low-priced-plans for its just-released flagship model, intensifying competition across a Chinese AI industry trying to take on Silicon Valley’s best https://t.co/fatPtQZPN2 ...

3) DeepSeek Sparse Attention (DSA)DeepSeek’s recently released V3.2 model introduced DeepSeek Sparse Attention (DSA), which brought complexity down from O(L²) to O(Lk), where k is fixed.How it works:A lightweight Lightning Indexer scores which tokens actually matter for each query.Small number of heads, runs in FP8, computationally cheap.Then a selection mechanism retrieves only the top-k key-value entries.The key insight is that only 2048 tokens get selected per query, regardless of context length.So the e ...

DeepSeek’s delayed release of its V4 model points to a strategic shift toward deeper integration with China’s domestic chip ecosystem, according to a social media account affiliated with the government-controlled CCTV https://t.co/dWyDBwpcoZ ...

Deepseek entered the AI race with a bang last year. Now it's back with a brand new model. The Chinese startup has just unveiled its V4 Flash and V4 Pro models.Models it says are the most powerful opensource AI models in the world. Posing a challenge arguably to competitors ranging from anthropic to open AI. They're claiming top tier coding performance, major leaps in reasoning, and more advanced agent style capabilities, all powered by upgraded architecture and heavy optimization.There is a catch. Capacity ...

DeepSeek has unveiled preview versions of a long-awaited new flagship model, which costs less than many alternatives to use but doesn’t meaningfully narrow the US lead in AI capabilities https://t.co/aP4esdZcUs ...