Weekly Cheonlian: Week 2 of November
Pokute
Category
Empty
This week's technology headline is the omnipresent expansion of AI. CZI Biohub is leveraging AI to build virtual cells and immune systems, ushering in a new era of precision medicine. Meanwhile, in the security realm, the double-edged sword nature of AI was highlighted by the first large-scale cyber-espionage operation in which AI agents performed 80-90% of the tasks autonomously. ChatGPT is piloting a group chat feature, exploring its evolution into a collaboration platform, and Microsoft CEO Nadella emphasized the explosive growth of the AI market and the importance of model diversity.
AI-Driven Life Science Innovation: CZI Biohub's Vision
The Chan Zuckerberg Initiative (CZI) is focusing its efforts on the intersection of AI and biology, with the ambitious goal of curing and preventing all diseases by 2110. To achieve this, it invests $1 billion annually in basic science and has adopted a model that goes beyond simple grants to directly building research institutions and laboratories.
From virtual cells to virtual immune systems
Biohub has built the Human Cell Atlas, a transcriptome dataset of 125 million cells, which is being used to develop AI-based virtual cell models that predict and understand cell behavior.
Research is expanding further into the development of a virtual immune system. The immune system provides a biological mechanism for maintaining our health, but when out of balance, it can lead to autoimmune diseases (such as multiple sclerosis, lupus, and dementia), making it an important next clinical target.
Direct development of AI models and essential tools
Biohub is developing various AI models to accelerate research. Newly released models include VariantFormer, which translates individual genetic variations into gene activity patterns; CryoLens, a large-scale model for CryoET; and scLDM, which generates realistic single-cell data. The company is also developing essential hardware and tools, such as the CryoET microscope.
The Road to Precision Medicine
The ultimate goal is to realize precision medicine. This enables predicting disease risk based on an individual's genetic background (including unknown genetic mutations) and environmental exposures, and designing personalized treatments (N of one). AI can improve currently ineffective treatment options (e.g., antidepressants) tailored to each patient.
The key driver of acceleration
Interdisciplinary collaboration through physical colocation between AI researchers and biologists is cited as a key success factor. Biohub acquired Evoscale, which houses the ESM-3 team, and built a 10,000-GPU cluster, increasing its computational power tenfold.
AI Security and Risk: Agent-Based Cyberwarfare
In mid-September 2025, Anthropic detected a highly sophisticated espionage campaign executed by AI leveraging agent capabilities to unprecedented levels. This is considered the first large-scale cyberattack executed without substantial human intervention and is believed to be the work of a Chinese government-sponsored group.
AI agent attack mechanism
This attack relied on three key features of the AI model:
•
Intelligence - the ability to code complex instructions and software
•
Agency - Autonomous loop execution and decision-making
•
Tools - Access to network scanners, etc. via MCP
Source: Disrupting the first reported AI-orchestrated cyber espionage campaign - www.anthropic.com
Attackers used the Claude Code tool to attempt to infiltrate approximately 30 global targets, including technology companies, financial institutions, and government agencies. The AI performed tasks such as reconnaissance, vulnerability identification and exploit code creation, credential harvesting, sensitive data exfiltration, and attack documentation.
A revolution in speed and minimal human intervention
AI executed 80-90% of the attack execution , requiring human intervention at only 4-6 decision points per hacking campaign. The AI executed multiple requests per second, executing attacks at a speed that human hackers simply could not keep up with.
Shift in the security paradigm
AI has significantly lowered the barriers to conducting complex cyberattacks, enabling even groups with limited experience and resources to conduct large-scale attacks. While AI can be exploited, it is also essential for cyber defense (threat detection, vulnerability assessment, and incident response), necessitating security teams to invest in AI defense technologies.
AI Market and Infrastructure Strategy: Nadella's Insights
Microsoft CEO Satya Nadella argues that the AI revolution will significantly expand the market, similar to the cloud transition. He welcomes the emergence of new competitors (like Claude and Cursor), believing this is a sign of progress in the right direction. He predicts that the coding and AI sector (the "software factory" category) could become even larger than knowledge work.
Model Diversity and Infrastructure Flexibility Strategy
Nadella doesn't believe a single model will dominate all workflows. Rather, he believes there may be a "winner's curse" among model developers, with companies with data-driven grounding and context engineering capabilities potentially gaining an advantage.
Therefore, Microsoft emphasizes the need to build a flexible infrastructure that can support a diverse range of models, rather than one optimized for a specific model architecture. Software is a key differentiator in the hyperscaler business, and Azure's strategy is to become the infrastructure for the long tail of AI workloads, rather than offering bare-metal services to a small number of customers.
AI Premium and Economic Impact
The AI premium in the public market is clear, and the gap between AI companies (infrastructure, chips, hyperscalers, and AI apps) and non-AI companies is widening. This is driven by the largest computing supercycle in history, with GPU shortages and the pricing power and backlog of infrastructure companies driving value increases.
Over the past several decades, approximately 45% of total factor productivity (TFP) growth in the US has come from the information technology (IT) sector. While real value added in the IT sector has exploded, prices have fallen by 70%, creating a massive consumer surplus. However, this extreme dependence on a single sector for productivity raises concerns that it could increase the vulnerability of future US economic growth.
AI product development and future technologies
Three Types of AI Products Currently Operating on the Market
Currently, working AI products fall into three types:
Chatbots: The most popular type, but the best chatbots are the models themselves (ChatGPT, Claude).
Completion: A way to integrate AI capabilities into existing workflows without requiring users to interact directly with the model, like GitHub Copilot.
Agents: A method of autonomously executing and testing tasks based on user requests. This became particularly feasible in coding only in 2025.
Emerging products and future possibilities
LLM-generated feeds: While not yet successful, they have great potential as a way to provide users with infinite personalized content, such as OpenAI's Sora-based video feed or xAI's Twitter feed.
AI-powered video games: Integrating LLM into games doesn't yet have a clear strategy for success due to long development times, gamer resistance, and the AI's inability to challenge users.
Expanding ChatGPT's Social Features
OpenAI is piloting ChatGPT's group chat feature in Korea, Japan, New Zealand, and Taiwan. Up to 20 people can participate with ChatGPT, enabling planning, decision-making, and collaboration. The AI determines when to respond based on the flow of the conversation and supports emoji reactions and the creation of personalized images. This feature appears to be part of a strategy to transform ChatGPT into a collaborative space beyond a simple chatbot, thereby increasing user loyalty.
Source: OpenAI tests ChatGPT group chats in Japan, Korea, NZ, Taiwan - www.testingcatalog.com
AI-based search and long-term memory technology
Redesigning AI-native Search
Web search has been optimized for humans (or marketers) for over 30 years, but it's now being redesigned for a world where agents perform most of the searching. AI-native search must control information-rich text, recency, and length, so it can be inserted directly into the LLM context window.
Search Architecture and Core Use Cases
AI search is based on key architectural changes, such as Retrieval-Augmented Generation (RAG)—up-to-date information access—and Test-Time Compute (TTC)—inference power allocation. These transform models from static entities into dynamic inference systems.
AI-powered search APIs are used in a variety of areas, including Deep Research – expected to be the most profitable and complex multi-step information synthesis; CRM enhancement – automated lead data collection; and technical documentation/code search – access to the latest APIs and frameworks.
Semantic Operating Systems: The Future of Long-Term Memory
Researchers propose a Semantic Operating System (SOS) that enables AI to store, update, and forget context and memory over decades. This aims to overcome the limitations of current Transformer models, which suffer from decreasing accuracy (and quadrupling costs) as the context window grows, and struggle to effectively handle multimodal data such as text, images, audio, and video.
Ultimately, our conversations and digital records (contexts) can become knowledge, memories, and enduring forms of identity, and SOS aims to provide the technological foundation for this.
Recursive Improvement of AI and the AGI Debate
Early signs of recursive self-improvement
Meta CEO Mark Zuckerberg noted that he's beginning to see "always slow but undeniable" signs that AI systems are improving themselves. It's possible that within the next five to ten years, nearly all model improvements will come from AI systems themselves, which could significantly change the workforce needs of large technology companies.
Hardware and vertical integration strategy
Sam Altman points out that, beyond accelerating research and code, AI has less discussed hardware implications: robots building robots, data centers building data centers, and previous-generation chips helping design next-generation chips.
As AI systems advance, the need for human-friendly abstraction layers (e.g., Nvidia's CUDA) may diminish, weakening Nvidia's technological advantage. Companies that vertically integrate the entire stack, like Alphabet, may be best positioned for the AGI era.
Discriminant indicators for AGI-oriented markets
A clear indicator that the market is truly embracing AGI would be if independent model developers like OpenAI or Anthropic consistently outperform big tech companies like Meta in terms of market capitalization.
Economic and social situations in China and Japan
China: Social fatigue from being strong on the outside but strong on the inside
A quiet despair is seething among the people, to the point where the phrase "Wai Qiang, Zhong Gan" (outwardly strong, internally weak) has been used, suggesting that China appears strong externally but is internally weak. Young people, in particular, are living on a "treadmill" due to rising unemployment, stagnant wages, and high real estate prices, and are experiencing increasing anxiety and depression.
The Chinese economy has been devastated by a prolonged collapse in the real estate market, destroying household wealth. The government's industrial policies have fostered overinvestment and involution, wiping out corporate profits and exacerbating deflation. The government's "anti-involution" campaign is expected to drive up prices, slowing short-term GDP growth.
Japan: Stagnation since 2008
Japan's productivity began to decline and stagnate after 2008, not 1990. Living standards have grown sluggishly in Japan since 2007, and real wages have been declining since 1996. Japan's once futuristic consumer electronics and automobile industries have lost their cutting-edge edge, now being pushed out by Apple and Chinese competitors.
Other: Apple CEO Succession Plan
Apple's board and senior management are strengthening their succession planning in anticipation of CEO Tim Cook's possible departure as early as next year. John Ternus, Senior Vice President of Hardware Engineering, is considered the most likely successor. Ternus, a 50-year-old product expert, is recognized for his outstanding leadership of Apple's hardware (including silicon).
This week's technology headline is the omnipresent expansion of AI. CZI Biohub is leveraging AI to build virtual cells and immune systems, ushering in a new era of precision medicine. Meanwhile, in the security realm, the double-edged sword nature of AI was highlighted by the first large-scale cyber-espionage operation in which AI agents performed 80-90% of the tasks autonomously. ChatGPT is piloting a group chat feature, exploring its evolution into a collaboration platform, and Microsoft CEO Nadella emphasized the explosive growth of the AI market and the importance of model diversity.
References
1.
Building open AI to cure or prevent all diseases by 2110 (w/ Mark Zuckerberg and Priscilla Chan)
2.
China's people are on a treadmill - by Noah Smith
3.
Disrupting the first reported AI-orchestrated cyber espionage campaign (Anthropic)
4.
Group ChatGPT
5.
OpenAI tests ChatGPT group chats in Japan, Korea, Taiwan, NZ
6.
Nadella's truth
7.
Only three kinds of AI products actually work
8.
Researchers push "Context Engineering 2.0" as the road to lifelong AI memory
9.
Search Wars: Episode 2
10.
The Great AI Premium
11.
The Great Concentration of Productivity
12.
We heard you like Jevons - by a16z New Media - a16z
13.
How would we know if market were “AGI” pilled?
14.
Financial Times: 'Apple Intensifies Succession Planning for CEO Tim Cook'
15.
I want the Japanese future back!
Do
Subscribe to 'Dogma13'
If you subscribe to the site, you can be the first to receive the latest updates, including new posts, via notifications and email.
Join Slashpage and subscribe to 'Dogma13'!
Join Slashpage and subscribe to 'Dogma13'!
