# Open Source的所有文章

在 HTX 新聞中心流覽與「Open Source」相關的最新資訊與深度分析。潘蓋市場趨勢、專案動態、技術進展及監管政策，提供權威的加密行業洞察。

Behind the MiniMax 'Pseudo-Open Source' Controversy: Has Yan Junjie's Ideal Succumbed to Capital Anxiety?

The article discusses the controversy surrounding MiniMax's "pseudo-open-source" release of its flagship M2.7 model. Initially presented as open-source, the model was later released under a "Modified-MIT" license that restricts commercial use without written permission, sparking backlash from the developer community. This move is seen as a departure from true open-source principles and reflects the tension between technological idealism and commercial pressures. MiniMax, which recently went public, faces significant financial challenges, with substantial losses despite growing revenues. The company’s decision to limit commercial use is interpreted as an effort to protect its revenue streams and control how its models are deployed, especially after instances of third-party misuse affected its reputation. The incident highlights a broader industry trend where "open-weight" models are increasingly distinguished from fully open-source ones, raising concerns about trust and legal certainty for developers and enterprises. The author concludes that MiniMax’s shift represents a pragmatic, if controversial, business strategy aimed at ensuring profitability, even at the cost of community trust.

marsbit1 小時前

Behind the MiniMax 'Pseudo-Open Source' Controversy: Has Yan Junjie's Ideal Succumbed to Capital Anxiety?

marsbit1 小時前

Can Humans Control AI? Anthropic Conducted an Experiment Using Qwen

Can Humans Control Superintelligent AI? Anthropic’s Experiment with Qwen Models Anthropic conducted an experiment to explore whether humans can supervise AI systems smarter than themselves—a core challenge in AI safety known as scalable oversight. The study simulated a “weak human overseer” using a small model (Qwen1.5-0.5B-Chat) and a “strong AI” using a more powerful model (Qwen3-4B-Base). The goal was to see if the strong model could learn effectively despite imperfect supervision. The key metric was Performance Gap Recovered (PGR). A PGR of 1 means the strong model reached its full potential, while 0 means it was limited by the weak supervisor. Initially, human researchers achieved a PGR of 0.23 after a week of work. Then, nine AI agents (Automated Alignment Researchers, or AARs) based on Claude Opus took over. In five days, they improved PGR to 0.97 through iterative experimentation—proposing ideas, coding, training, and analyzing results. The findings suggest that, in well-defined and automatically scorable tasks, AI can help overcome the supervision gap. However, the methods didn’t generalize perfectly to unseen tasks, and applying them to a production model like Claude Sonnet didn’t yield significant improvements. The study highlights that while AI can automate parts of alignment research, human oversight remains essential to prevent “gaming” of evaluation systems and to handle more complex, real-world problems. Anthropic chose Qwen models for their open-source nature, performance, scalability, and reproducibility—key for rigorous and repeatable experiments. The research demonstrates progress toward automated alignment tools but also underscores that AI supervision remains a nuanced, human-AI collaborative effort.

marsbit昨天 09:28

Can Humans Control AI? Anthropic Conducted an Experiment Using Qwen

marsbit昨天 09:28

Someone Turned Buffett and Munger into Agents, Then Open-Sourced It...

The open-source project "AI Hedge Fund" has gained significant traction on GitHub, creating AI agents modeled after 12 legendary investors, including Warren Buffett and Charlie Munger, to analyze stocks and develop trading strategies. It features a team of 6 analyst agents that synthesize insights and make final decisions. The system includes a backtesting module to evaluate strategies with historical data before real investment. Built by developer Virat Singh, the project supports 13 major AI models and can run locally. It uses a React front-end with a visual workflow editor and a Python/FastAPI backend orchestrated with LangGraph. The agent team covers diverse philosophies, from value investing (Graham) to growth (Cathie Wood) and risk management (Taleb). While not proven in live markets, it offers a platform for learning agent frameworks and diverse investment perspectives through simulated expert debate. The project highlights a growing trend of "distilling" financial wisdom into AI, but users are cautioned about its unverified returns and investment risks.

marsbit前天 11:16

Someone Turned Buffett and Munger into Agents, Then Open-Sourced It...

marsbit前天 11:16

Hermes Agent Guide: Surpassing OpenClaw, Boosting Productivity by 100x

A guide to Hermes Agent, an open-source AI agent framework by Nous Research, positioned as a powerful alternative to OpenClaw. It is described as a self-evolving agent with a built-in learning loop that autonomously creates skills from experience, continuously improves them, and solidifies knowledge into reusable assets. Its core features include a memory system (storing environment info and user preferences in MEMORY.md and USER.md) and a skill system that generates structured documentation for complex tasks. The agent boasts over 40 built-in tools for web search, browser automation, vision, image generation, and text-to-speech. It supports scheduling automated tasks and can run on various infrastructures, from a $5 VPS to GPU clusters. Popular tools within its ecosystem include the Hindsight memory plugin, the Anthropic Cybersecurity Skills pack, and the mission-control dashboard for agent orchestration. Key differentiators from OpenClaw are its architecture philosophy—centered on the agent's own execution loop rather than a central controller—and its autonomous skill generation versus OpenClaw's manually written skills. Installation is a one-line command, and setup is guided. It integrates with messaging platforms like Telegram, Discord, and Slack. It's suited for scenarios requiring a persistent, context-aware assistant that improves over time, automates workflows, and operates across various deployment environments.

marsbit前天 13:11

Hermes Agent Guide: Surpassing OpenClaw, Boosting Productivity by 100x

marsbit前天 13:11

The Creator of Kling Returns to Alibaba and Builds Another Dark Horse

The article discusses the rise of HappyHorse-1.0, an AI video generation model developed by Alibaba, which topped the Artificial Analysis leaderboard in both text-to-video and image-to-video categories in April 2026. The model was created under the leadership of Zhang Di, who returned to Alibaba in November 2025 after working at Kuaishou, where he led the development of the Kling model. HappyHorse is open-source and commercially available, similar to Alibaba's Qwen model. Zhang Di's background includes extensive experience in large-scale data systems and machine learning at Alibaba and Kuaishou, which contributed to the rapid development of HappyHorse within just five months. The model uses a 15-billion-parameter transformer architecture with native multimodal training, supporting multiple languages and lip-sync capabilities. It also focuses on reducing inference time and cost, making it practical for commercial use. The primary application of HappyHorse is in e-commerce, where it can generate product videos to enhance user engagement and conversion rates by creating contextual and personalized content. This aligns with Alibaba's strengths in commerce, advertising, and data feedback loops. The model's success with open-source approach contrasts with challenges faced by closed-source models like OpenAI's Sora (shut down due to high costs) and ByteDance's Seedance 2.0 (paused over copyright issues). HappyHorse represents a strategic move for Alibaba to integrate AI video generation into its core business ecosystems.

marsbit04/13 05:10

The Creator of Kling Returns to Alibaba and Builds Another Dark Horse

marsbit04/13 05:10

When AI's Bottleneck Is No Longer the Model: Perseus Yang's Open Source Ecosystem Building Practices and Reflections

In 2026, the AI industry's primary bottleneck is no longer model capability but rather the encoding of domain knowledge, agent-world interfaces, and toolchain maturity. The open-source community is rapidly bridging this gap, evidenced by projects like OpenClaw and Claude Code experiencing explosive growth in their Skill ecosystems. Perseus Yang, a contributor to over a dozen AI open-source projects, argues that Skill systems are the most underestimated infrastructure of the AI agent era. They enable non-coders to program AI by writing natural language SKILL.md files, transferring power from engineers to all professionals. His project, GTM Engineer Skills, demonstrates this by automating go-to-market workflows, proving Skills can extend far beyond engineering into areas like product strategy and business analysis. He also identifies a critical blind spot: while browser automation thrives, agent operations are nearly absent from mobile apps, the world's dominant computing interface. His project, OpenPocket, is an open-source framework that allows agents to operate Android devices via ADB. It features human-in-the-loop security, agent isolation, and the ability for agents to autonomously create and save new reusable Skills. Yang believes the value of open source lies not in the code itself, but in defining the infrastructure standards during this formative period. His work validates the SKILL.md format as a portable unit for agent capability and pioneers new architectures for agent operation in API-less environments. His design philosophy prioritizes usability for non-technical users, ensuring the agent ecosystem can be expanded by practitioners from all fields, not just engineers.

marsbit04/13 01:29

When AI's Bottleneck Is No Longer the Model: Perseus Yang's Open Source Ecosystem Building Practices and Reflections

marsbit04/13 01:29

Anthropic's 'Cry' Sparks Wall Street Panic! 27-Year-Old, Mythos Instantly Defeated by 8 AIs

Anthropic's announcement of Claude Mythos, an AI tool claiming to discover thousands of zero-day vulnerabilities—including a 27-year-old bug in OpenBSD—triggered alarm on Wall Street and prompted emergency meetings among financial regulators fearing systemic cyberattacks. However, independent tests revealed significant exaggerations in these claims. Researchers found that many reported vulnerabilities were in obsolete software or were impractical to exploit, and the findings relied on only 198 manual reviews. Furthermore, multiple smaller, open-source AI models (some with as few as 3 billion parameters) successfully identified the same critical flaws at a fraction of the cost, demonstrating that AI cybersecurity capability does not linearly scale with model size. Meanwhile, users reported severe performance degradation in Claude Opus 4.6, with reduced reasoning depth and increased API costs. Critics, including prominent hacker George Hotz, accused Anthropic of overstating risks for marketing purposes, creating a "wolf cry" scenario where hype overshadows reality.

marsbit04/12 08:03

Anthropic's 'Cry' Sparks Wall Street Panic! 27-Year-Old, Mythos Instantly Defeated by 8 AIs

marsbit04/12 08:03

DeepMind Founder's Latest Interview: AGI May Be Achieved Within 5 Years, Scale 10 Times That of the Industrial Revolution, Previous Wave of Ideas Has Been 'Squeezed Dry'

DeepMind co-founder Demis Hassabis predicts that AGI, defined as a system with all cognitive abilities of the human mind, has a high probability of being realized within the next five years. He identifies compute power as the primary bottleneck, though scaling laws continue to yield significant returns. Key missing capabilities include continuous learning and robust generalization, which he terms "jagged intelligence." Hassabis believes AGI will become the most powerful tool in science and medicine, potentially leading to a golden age of discovery. He emphasizes the need for global coordination on AI safety and regulation, proposing an international framework similar to the IAEA. While acknowledging risks like misuse and labor disruption, he views the AI revolution as potentially ten times the scale and speed of the Industrial Revolution. He advocates for equitable distribution of benefits and sees AI driving breakthroughs in energy, materials science, and healthcare, including cancer cures.

marsbit04/10 00:00

DeepMind Founder's Latest Interview: AGI May Be Achieved Within 5 Years, Scale 10 Times That of the Industrial Revolution, Previous Wave of Ideas Has Been 'Squeezed Dry'

marsbit04/10 00:00

Mysterious Model HappyHorse Tops the Chart Overnight: Is the Video Generation Arena Welcoming a "Game Changer"?

A mysterious AI video generation model named "HappyHorse-1.0" has quietly topped the AI Video Arena leaderboard on Artificial Analysis, surpassing established models like Seedance 2.0 and others in Elo score—a user-blind-test-based ranking reflecting real perceived quality. The model’s origin was initially unknown, but technical analysis later linked it to the open-source model "daVinci-MagiHuman," jointly developed by Shanghai SII GAIR Lab and Beijing-based Sand.ai. HappyHorse-1.0, likely an optimized iteration by Sand.ai, uses a 15-billion-parameter transformer architecture for joint audio-video-text modeling. Its strong performance in human-centric scenes (e.g., portraits, narrations) helped it excel in blind tests, though it still lags in multi-character or complex motion scenarios. The achievement signals a potential shift: an open-source model rivaling closed-source alternatives in perceived quality, which could lower costs and increase flexibility for developers in vertical applications like virtual avatars. However, limitations remain, including high computational requirements (H100 GPU needed) and shorter generation lengths. While not yet threatening market leaders, HappyHorse represents progress toward open models reaching "production-ready" quality, potentially accelerating community-driven improvements in the video AI space.

marsbit04/08 07:57

Mysterious Model HappyHorse Tops the Chart Overnight: Is the Video Generation Arena Welcoming a "Game Changer"?

marsbit04/08 07:57

Running Gemma 4 Locally on iPhone Goes Viral: How Far Are We from the Zero Token Era?

Google's newly open-sourced Gemma 4 model, built on the same architecture as Gemini 3, has gained significant attention for its ability to run locally on mobile devices like the iPhone and Samsung Galaxy. With smaller versions such as E2B (2.3B parameters) and E4B (4.5B parameters), it supports native multimodal capabilities and offers a 128K context window. Users report impressive speeds—over 40 tokens per second on Apple chips with MLX optimization—making it feel "like magic." The model is accessible via Google’s official AI Edge Gallery app, ensuring ease of use and security. While Gemma 4 excels in tasks like text generation, coding, and image understanding, it struggles with more complex agent-based workflows, such as tool calling and structured outputs, where models like Qwen3-coder perform better. Despite some limitations in reasoning, Gemma 4’s local performance hints at a future where everyday AI tasks—chat, coding, reasoning—can be handled offline, reducing reliance on cloud-based token services. Although cloud models still lead in advanced reasoning and large-scale multi-agent tasks, the trend suggests that as hardware and quantization improve, on-device models will increasingly handle high-frequency simple tasks. This shift could disrupt the AI industry’s reliance on token sales and API subscriptions, pushing providers to focus on more complex, data-intensive capabilities. Gemma 4 is just the beginning of this transformation.

marsbit04/06 05:53

Running Gemma 4 Locally on iPhone Goes Viral: How Far Are We from the Zero Token Era?

marsbit04/06 05:53

活动图片

熱門分類right-arrow

熱門標籤right-arrow

Market Analysis