Author: RadarAI Editorial
Editor: RadarAI Editorial
Last updated: 2026-07-07
Review status: Editorial review pending
Brief
速报
官方
AI动态
开源
Global AI competition is accelerating deeper into hardware supply chains, interpretable architectures, and finance-specific vertical applications; Tencent shifts strategy—selling over RMB 10 billion worth of Kuaishou shares while increasing investments in Keling AI and DeepSeek. Meanwhile, Anthropic's J-space mechanism reveals that less than 10% of large model neural activity carries accessible information [8], and Wall Street institutions are collectively reallocating assets to bet on Chinese AI chipmakers and cost-effective large model ecosystems [1].
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
Global AI competition is rapidly deepening into **hardware supply chains**, **interpretable architectures**, and **finance-specific vertical applications**; **Tencent's strategic pivot**—selling over RMB 10 billion worth of Kuaishou shares while ramping up investments in Keling AI and DeepSeek—coincides with **Anthropic's** revelation via its J-space mechanism that only under 10% of a large model's neural activity carries accessible information [8]; meanwhile, Wall Street institutions are collectively rebalancing portfolios to back **Chinese AI chip companies** and high-value-per-dollar large model ecosystems [1].
## 🚀 Key Developments
- **Tencent sells over RMB 10 billion in Kuaishou shares, exiting as a major shareholder—and simultaneously boosts investment in Keling AI and DeepSeek** [4]: The divestment netted over HK$10 billion, reducing Tencent's stake to 9.37%, widely interpreted as a strategic reallocation of AI assets.
- **Anthropic unveils Claude's internal 'J-space' interpretability mechanism** [8]: Only ~10% of neural activity encodes accessible information—supporting multi-step reasoning and safe behavior—but this does not equate to subjective consciousness.
- **Wall Street institutions collectively pivot toward China's AI sector** [1]: Macquarie highlights strong growth potential in domestic AI chipmakers; price disparities between U.S. and Chinese large model services reach tens of times—making local supply chains and cost efficiency new focal points for capital.
- **Claude-Video framework enables video-level multimodal understanding** [13]: Integrates yt-dlp for downloading, intelligent frame sampling, deduplication, and Whisper-based subtitle transcription—supporting configurable frame budgets and adaptive FPS.
- **The Karpathy Loop boosts AI experimentation efficiency by 10×** [12]: Implements an automated, iterative agent workflow—replacing manual, one-off prompting—to significantly increase R&D throughput.
- **Shepherd framework introduces 'Git-style' version control for agent execution** [15]: Supports forking, rollback, and replay—with 95% KV cache reuse—enabling supervised training and agent self-optimization.
- **AI deployment hits hardware supply chain bottlenecks** [7]: Small teams developing acoustic robots face four core challenges: component shortages, inconsistent quality, lack of customization capability, and low vendor priority—highlighting gaps in domestic supply chain coordination.
- **Former OpenAI CTO's team fine-tunes Qwen3-235B into a financial domain expert model** [3]: Trained via the Tinker platform, it outperforms GPT-4 and Gemini Pro on financial information filtering tasks.
## 🔗 Sources
[1] Wall Street Institutions Bet Big on China's AI — https://www.bestblogs.dev/article/1ff2b65f?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[3] 'Mother of ChatGPT' Fine-Tunes Qwen Model—Outperforms GPT & Gemini — https://www.bestblogs.dev/article/d5100cdb?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[4] Tencent Sells Over RMB 10 Billion in Kuaishou Shares — https://www.bestblogs.dev/article/cc4c4c51?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[7] AI Deployment Still Gets Schooled by Hardware Supply Chains — https://www.bestblogs.dev/article/4f05e5cd?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[8] Anthropic Reveals J-Space in Claude: An Interpretable Mechanism Analogous to Global Workspace Theory — https://www.bestblogs.dev/status/2074317616894955582?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[12] Karpathy Loop
Global AI competition is rapidly deepening into hardware supply chains, interpretable architectures, and finance-specific vertical applications; Tencent's strategic pivot—selling over RMB 10 billion worth of Kuaishou shares while ramping up investments in Keling AI and DeepSeek—coincides with Anthropic's revelation via its J-space mechanism that only under 10% of a large model's neural activity carries accessible information [8]; meanwhile, Wall Street institutions are collectively rebalancing portfolios to back Chinese AI chip companies and high-value-per-dollar large model ecosystems [1].
🚀 Key Developments
- Tencent sells over RMB 10 billion in Kuaishou shares, exiting as a major shareholder—and simultaneously boosts investment in Keling AI and DeepSeek [4]: The divestment netted over HK$10 billion, reducing Tencent's stake to 9.37%, widely interpreted as a strategic reallocation of AI assets.
- Anthropic unveils Claude's internal 'J-space' interpretability mechanism [8]: Only ~10% of neural activity encodes accessible information—supporting multi-step reasoning and safe behavior—but this does not equate to subjective consciousness.
- Wall Street institutions collectively pivot toward China's AI sector [1]: Macquarie highlights strong growth potential in domestic AI chipmakers; price disparities between U.S. and Chinese large model services reach tens of times—making local supply chains and cost efficiency new focal points for capital.
- Claude-Video framework enables video-level multimodal understanding [13]: Integrates yt-dlp for downloading, intelligent frame sampling, deduplication, and Whisper-based subtitle transcription—supporting configurable frame budgets and adaptive FPS.
- The Karpathy Loop boosts AI experimentation efficiency by 10× [12]: Implements an automated, iterative agent workflow—replacing manual, one-off prompting—to significantly increase R&D throughput.
- Shepherd framework introduces 'Git-style' version control for agent execution [15]: Supports forking, rollback, and replay—with 95% KV cache reuse—enabling supervised training and agent self-optimization.
- AI deployment hits hardware supply chain bottlenecks [7]: Small teams developing acoustic robots face four core challenges: component shortages, inconsistent quality, lack of customization capability, and low vendor priority—highlighting gaps in domestic supply chain coordination.
- Former OpenAI CTO's team fine-tunes Qwen3-235B into a financial domain expert model [3]: Trained via the Tinker platform, it outperforms GPT-4 and Gemini Pro on financial information filtering tasks.
🔗 Sources
[1] Wall Street Institutions Bet Big on China's AI — https://www.bestblogs.dev/article/1ff2b65f?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[3] 'Mother of ChatGPT' Fine-Tunes Qwen Model—Outperforms GPT & Gemini — https://www.bestblogs.dev/article/d5100cdb?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[4] Tencent Sells Over RMB 10 Billion in Kuaishou Shares — https://www.bestblogs.dev/article/cc4c4c51?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[7] AI Deployment Still Gets Schooled by Hardware Supply Chains — https://www.bestblogs.dev/article/4f05e5cd?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[8] Anthropic Reveals J-Space in Claude: An Interpretable Mechanism Analogous to Global Workspace Theory — https://www.bestblogs.dev/status/2074317616894955582?utm_source=rss&utm_medium=feed&utm_campaign=resources&entry=rss_article_item
[12] Karpathy Loop
← Back to Updates