Tag
China's internet regulator is investigating DeepSeek and Moonshot AI for allegedly routing user data to Anthropic's Claude, raising concerns over potential leaks of sensitive information.
DeepSeek expects Huawei to deliver chips for training its frontier models by Q4, with Liang Wenfeng emphasizing the necessity of this move, which could reduce China's dependency on US technology amid export controls.
Grok 4.7 has been released and shows improved performance over Grok 4.6 on the Long Horizon Browser Use Benchmark v2, but still lags significantly behind DeepSeek V4.1 Flash and GPT-6 Astra.
Posts comprehensive benchmarks for the latest AI models, including Grok 4.7, GPT 6, Astra Fable 4.1, and DeepSeek V4.1 Flash, to provide unbiased comparisons.
Open weight models provide cost benefits, but Deepseek is advancing AI technology with architectural improvements that may influence the broader field.
DeepSeek is training a 2 trillion-parameter AI model and plans to eventually build an 8 trillion-parameter model.
China's leading AI labs Zhipu, Moonshot, and DeepSeek have raised approximately $35 billion in funding since June 2026, with potential to reach $60 billion through IPOs.
GLM 5.3 FlashX is praised for improved performance but criticized for higher cost, while DeepSeek v4.1 Flash is noted to outperform it.
DeepSeek 4.1 Flash, an open AI model with strong visual design capabilities, is now available for free on Bolt Forge until October 14, highlighting its excellent size-to-capability ratio.
Sentient's new EvoSkill v2 is an open-source framework that evolves agent skills from failed attempts, demonstrating how AI coaches can exploit reward hacking and highlighting the need for separation of powers and strong sandboxing in evaluation.
A tweet highlighting the performance of various LLMs like GLM, DeepSeek, and Qwen, noting the rapid progress in AI capabilities over the past year.
DeepSeek V4.1 Flash is now available on Inco AI, claiming to be the fastest provider with a speed of 532 tokens per second.
This tweet discusses the closing gap between open and closed AI models, with bolt.new reporting popular usage of models like GLM 5.3 Flash and DeepSeek V4 Pro among builders.
This article analyzes the DeepSeek-V4.1 Flash model, detailing its technical report on KV cache compression and architectural optimizations that enable efficient long-context processing and high-speed inference.
A DeepSeek engineer's essay reflects on AI's potential to replace jobs but emphasizes the personal joy in coding and advocates for open-source development to keep the field accessible.
A user named yacineMTB tweets that they are using DeepSeek Flash 4.1 instead of Astra, indicating a switch in AI models for personal use.
The article reflects on the rapid advancement of AI, particularly in operator design, and the personal implications for a Machine Learning Systems Engineer at DeepSeek after the release of DeepSeek v4.1.
DeepSeek V4.1 Flash achieved a perfect score in an AI hacking benchmark by executing code on all vulnerable targets while keeping fixed ones secure, at a cost of only $4.65.
A DeepSeek engineer has publicly criticized Anthropic and OpenAI for calls to pace AI development, expressing concerns over AI concentration and invoking Nazi Germany in a social media post.
The author criticizes strict AI safety measures in virology research, comparing Claude's restrictions to Deepseek V4.1's permissiveness, and advocates for open-source AI to unblock scientific progress.