Tag
Viggle has released the JST-2 model, which uses a "physics-native" Joint Space-and-Time architecture to represent the world as 3D atoms with spatial positions and physical properties. It can generate interactive 3D scenes, skeletal characters, and full-body motions in about one second for less than a cent. The model is available via API but is not yet open-sourced.
Gergely Orosz 反应 Anthropic 工程博客关于自研源代码管理系统的文章，指出该项目和 Anthropic 其他产品一样是闭源的，只有 Bun 和 MCP 是开放的。原推文提到 Anthropic 每日代码合并量从 2,500 涨至 30,000+，迫使其为 agent 规模重建开发基础设施。
A reverse-engineering repo (REA) could let AI agents systematically decompile and reconstruct closed-source software, potentially feeding AI labs like DeepSeek and Z ai with vast datasets of real engineering problems and accelerating coding capabilities across the industry.
The tweet argues that in modern tech, the open source vs. closed source distinction is less important due to reverse-engineering and replication, making core product launches competitive races and defensible niches rare.
An open-source AI model named Laya has quickly emerged to rival the closed-source Jev, outperforming it in some metrics after fine-tuning, underscoring the rapid progress in open-source development versus secretive startup efforts.
StepFun, a Chinese AI lab, has released Step 5 Preview, which matches the performance of Kimi K3 at one-third the cost, marking its entry into the AI frontier.
Jev is a closed-source AI model released by TypeSafe, designed for rapid judgment and selection tasks, featuring low latency and low cost. It is widely used in automated workflows and Agent systems.
The article argues that closed-source AI is more dangerous than open-source AI, emphasizing that security through obscurity is ineffective.
A tweet discussing issues in the open-source AI community, where researchers hedge their careers by aligning with closed-source labs, and referencing Anthropic's stance on user interactions.
A tweet suggesting the urgent reverse engineering of closed source software to preempt copyright-based restrictions.
The article argues that open source AI is catching up to closed source due to collaborative efforts among independent labs, especially in Chinese ones, and suggests major companies like OpenAI and Anthropic could improve by sharing knowledge.
The article discusses how the AI competition is evolving into a contrast between closed frontier intelligence and open-weight intelligence, which may have greater long-term implications for power distribution among AI companies like OpenAI, Google, and Meta.
Raphael Amorim announces he is closing the source code of Canario Terminal, citing the high maintenance cost of open source, AI-generated spam, and lack of time. He will continue maintaining Rio and notes Canario has no telemetry, with older source still available in Rio's history.
Jensen Huang was asked whether open-source model companies should be allowed to distill closed-source models. He said distillation is not copying, but a fundamental way of intelligence.
Discusses why, among iOS proxy tools, the paid and closed-source Shadowrocket is more popular than the free and open-source Clash Mi.
CNBC argues the US must shift from a hardware-centric AI strategy to an open-source AI strategy, as Chinese open-weight models gain global traction, warning of repeating the semiconductor manufacturing offshoring mistake.
The article discusses the shift in the AI race from model strength to ownership, highlighting the concept of 'AI communism' and the philosophical split between open and closed AI development, with references to figures like Dean Ball, Elon Musk, and Tang Jie.
A discussion questioning whether open-source models like Kimi K3 or GLM can replicate the mathematical and cybersecurity problem-solving achievements recently demonstrated by closed-source models from OpenAI and Anthropic.
This paper explores the feasibility of reconstructing the tokenizer of a closed-source large language model using only two oracle queries from its chat API, potentially revealing internal representation details.
A recent OpenRouter study of 100 trillion tokens provides high-quality empirical data comparing open and closed frontier model usage, now available on arXiv.