資本之眼從研究員手中接走 AGI 火炬(AI 生成示意圖)

[Weekly AGI Observation]: Karpathy Switches Sides, Maia Lends Troops, 900 Billion Valuation — Anthropic Completes Credit Upgrade in One Week — Issue 4

Within a single week, Anthropic received three credit upgrade certificates: Karpathy joined Claude pre-training, Microsoft opened Maia 200 for rent, and a new $30B round is set to close at a $900B valuation. At Google I/O 2026, Hassabis stated we are "at the foothills of the singularity"; Marc Andreessen claimed on Joe Rogan that AGI was achieved three months ago; yet ARC-AGI-3 remains at 100% for humans vs. 0.51% for the strongest AI. This week, the Xiaojian Index rose from 4.6 to 4.8.

遞迴自我改進變成創投資產類別(AI 生成示意圖)

[This Week in AGI]: When “Self-Improvement” Becomes an Investable Asset Class — Issue 3

What we saw this week wasn't a breakthrough demo, but Silicon Valley and Wall Street joining forces to turn "recursive self-improvement" from a paper paragraph into a check-writing asset class: Recursive Superintelligence emerged with a $4.65B valuation, Isomorphic Labs raised a $2.1B Series B, Claude Mythos hit the 16-hour ceiling on METR, and the White House shifted from sanctioning Anthropic to drafting an "Executive Order bypass". The AGI Realization Tracking Index rose from 4.2 to 4.6.

[This Week’s AGI Observation] Recursive Boot, Abandoning Definitions — Issue 2

Google AlphaEvolve has completed the "Improve Gemini → Gemini trains a stronger AlphaEvolve" closed loop in its data centers; Sam Altman dropped the "AGI may have already whooshed by" remark, pushing the abandonment of definitions to its full extent; Anthropic's valuation surged from $350 billion to over $900 billion in 4 weeks, with an IPO potentially coming in October. The Xiao Jian AGI Implementation Tracking Index rose from 3.9 to 4.2 this week.

AI 生成示意圖:兩個 AI 巨頭在天平上的權力轉移

The Pentagon’s ‘Eight Minus One’: Anthropic Refuses to Be a Tamed AI

On May 1st, the U.S. Department of Defense announced that eight companies—OpenAI, Google, Microsoft, AWS, Nvidia, SpaceX, Reflection, and Oracle—could enter IL6/IL7 classified networks, with only Anthropic missing. The reason is not a lack of capability, but Anthropic's refusal to accept the "any lawful purpose" clause, unwilling to hand over red lines regarding autonomous weapons and large-scale surveillance. In the same week, Anthropic secured $40 billion from Google but was listed as a "Supply Chain Risk" by the Pentagon, with the CTO stating on the same day that Mythos is a separate issue. This is not a contract dispute; it is a battle of paradigms.

AI 生成示意圖:兩個 AI 巨頭在天平上的權力轉移

OpenAI Misses, Anthropic Takes All: The Power Balance of the AGI Game is Tilting

The financial figures from the past week tell us more directly where the AGI game stands than any model benchmark score. For the first time, OpenAI has been reported by its own insiders to have missed internal revenue and user targets, with even the CFO disagreeing with the CEO in internal memos. In the same week, Anthropic secured $40 billion from Google, receiving backing from all four major tech giants. With the AGI clause removed from contracts, cash flow has become the real judge. Based on these figures, I discuss how this game is shifting.

Claude AI 2026 完整教學封面圖

Complete Claude AI Tutorial 2026: How to Use the Anthropic Chatbot? Free Features, Pro Plans, and Cowork Desktop Version Explained

If you have recently searched for the “most powerful AI chatbot” online, the name Claude AI has surely appeared more than once. Developed by Anthropic, Claude has evolved into its fourth generation by 2026, delivering impressive performance in writing articles, analyzing data, coding, and summarizing long reports. More importantly, Claude has a free version that you can use immediately without needing to pull out a credit card. This article will take you from zero to hero in understanding Claude AI: what features the free version offers, the differences between Pro and Team plans, a 5-step quick-start guide, and an exclusive…

AGI Clause 被埋葬,DeepSeek V4 與 Mythos 等紙條飄落 — 第 1 期週報封面(AI 生成示意圖)

[This Week’s AGI Observation]: The Week the “AGI Clause” Was Buried — Issue 1 / 2026-04-28

This week, OpenAI and Microsoft buried the "AGI termination of exclusivity" clause. In the same week, OpenAI rewrote its charter into the "AGI Five Principles," Anthropic locked its most powerful model, Mythos, into the private Project Glasswing alliance, and DeepSeek V4 ran 1.6T parameters using entirely domestic chips. Xiao Jian's Issue 1 deep dive: AGI has shifted from a "discrete event" to a "jagged process." This week's Xiao Jian Index: 3.9 / 10 (↑ 0.3).

InVideo AI 完整教學 2026 文字轉影片 AI 影片生成器

InVideo AI Complete Tutorial 2026: How to Use Text-to-Video? Free Version Features, Subscription Plans, and Advanced Tips at a Glance

InVideo AI is one of the most powerful AI video generators available today, capable of generating complete videos from simple text input. This article provides a comprehensive guide to InVideo AI's latest 2026 features, v4 Agent, free version limitations, a comparison of the three subscription plans, and a three-step quick-start tutorial!

Krea AI 完整教學 2026 - 創意 AI 套件

Krea AI Complete Tutorial 2026: AI Images, Videos, Real-time Generation, All in One Platform!

Have you ever wondered if there is an AI tool that allows you to see the effects of AI image generation in real-time without learning complex prompts, and can also generate videos, enhance image resolution, or even train your own AI models? The answer is: yes, and it’s called Krea AI. Krea AI is one of the fastest-growing AI creative suites in recent years, with monthly searches reaching the tens of millions and an excellent reputation in the AI drawing community. Its biggest highlight is the “Real-time Generation” feature—you draw a sketch, and it generates the image simultaneously, completely breaking…