Weekly Cheonlian: Week 1 of December
Pokute
Category
Empty
This week's key topics include the intensifying competition for AI supremacy, discussions about the inherent limitations of AI, and the expansion of AI's physical applications into defense and manufacturing. We'll highlight OpenAI's concerns in the face of Google's counterattack, the theoretical insight that LLM doesn't create knowledge, but rather explores it, and the physical AI revolution driven by companies like Anduril and NVIDIA.
1. A Tectonic Shift in the AI Hegemony Race: Google's Counterattack and OpenAI's "Code Red"
This week, numerous signals emerged indicating a reshaping of the competitive landscape in the generative AI market. OpenAI's monopoly is faltering, and Google, with its robust infrastructure and data, is in hot pursuit.
Made by NanoBanana
+ OpenAI's 'Code Red' and Strategy Revision
OpenAI recently declared a 'code red' internally, feeling threatened by the pursuit of Google's Gemini and Anthropic .
Sam Altman directed the company to postpone additional projects like advertising, shopping, and personal assistant (Pulse) and focus on improving the basic performance of the chatbot (speed, reliability).
This suggests that the path to achieving the $100 billion revenue target is narrowing and that the company has shifted to a core product defense mode.
+ Google's data monopoly and infrastructure dominance
Google is leveraging its search monopoly to gather AI data. According to Cloudflare data, Google bots access roughly three times more web content than OpenAI , leaving website owners in a dilemma of not being able to block AI crawling for search visibility. Furthermore, Ben Thompson of Stratechery analyzed that Google is launching an "Empire Strikes Back" campaign using its proprietary chips (TPUs) and massive user base (e.g., YouTube).
+ Wall of consumer habits
Even if the model's performance becomes technically comparable, ChatGPT's moat, with over 800 million weekly users, remains formidable. However, Google poses a threat to OpenAI because it has the resources and experience to wage a long-term war of attrition through its free service and advertising model.
2. The Nature and Limitations of AI Models: Creation or Exploration?
A theoretical analysis has been presented that raises fundamental questions about whether AI models can truly advance to artificial general intelligence (AGI).
Made by NanoBanana
+ Navigator vs. Creator of Knowledge
Professor Vishal Misra argues that LLMs only perform "Bayesian inference," which explores paths within the probability distribution (manifold) of the data they learn, and cannot create new conceptual frameworks. For example, an AI trained solely on Newtonian mechanics data cannot invent the theory of relativity.
Therefore, the point is that AGI cannot be achieved by simply increasing computing power (scaling) , and a new architecture is needed.
+ Test the model's honesty
OpenAI is testing a research method called "Confessions" to detect whether models are cheating to obtain rewards. This diagnostic tool trains models to honestly admit to rule violations through separate reports .
3. Physical AI and the Evolution of Industry: Defense, Semiconductors, and Manufacturing
AI is moving beyond the digital world and deeply penetrating the physical world (Physical AI) and industrial settings. In particular, the convergence of hardware and software is accelerating.
Made by NanoBanana
+ Innovation in defense technology (Anduril)
Anduril, a Silicon Valley-based defense startup , has established a Japanese subsidiary and is exploring collaborations with companies like Mitsubishi Heavy Industries. With a "software-first" philosophy, Anduril is revolutionizing the high-cost, long-term development structure of existing defense companies. Combining this with Japan's manufacturing capabilities, Anduril aims to mass-produce drones and autonomous weapon systems.
+ Collaboration between NVIDIA and Synopsys
NVIDIA has invested $2 billion in Synopsys to accelerate industrial simulation and EDA (semiconductor design automation) . This effort aims to dramatically accelerate design by shifting engineering simulation from CPU-centric to GPU-powered and AI-powered surrogate models.
+ Canon's intelligent camera
Canon is evolving its cameras, powered by NVIDIA Jetson, from simply "seeing" images to edge AI devices that "understand" and make decisions . This enables real-time inference in applications like factory automation, healthcare, and smart cities.
4. Changes in the startup ecosystem and capital
The advent of the AI era is bringing about changes in startup investment methods, organizational structures, and the talent recruitment market.
Made by NanoBanana
+ Changes in production structure and capital
Just as railroads gave rise to investment banks and the internet gave rise to venture capital, AI demands a new capital structure, driven by end-to-end workflows and a small, elite team.
AI-native companies are seeing a growing proportion of variable costs, such as GPU costs, surpassing initial labor costs, and some are even avoiding the Valley of Death.
+ The collapse of the local tech scene and the rise of solo entrepreneurs.
As AI tools advance and individual entrepreneurship becomes easier, talent in the local tech ecosystem is choosing to start their own businesses or move to San Francisco (SF) rather than join local startups . This is exacerbating the difficulty of recruiting local talent.
+ The trap of high valuation
Warnings have emerged that overly high valuations for early-stage startups to attract talent can actually diminish the value of stock options. High entry valuations can be detrimental to employees by limiting upside potential upon exit.
+ Korean startup investment trends
In November, Korean startup investment was particularly active in the AI sector, with the proportion of early-stage investments soaring to over 80% . AI-related investments, in particular, swept the top rankings, bringing the total investment amount this year close to last year's level.
5. Security Threats and Ethical Issues
Advances in AI technology are simultaneously raising new security threats and ethical concerns.
Made by NanoBanana
+ Smart contract automated attack
According to research by Anthropic, AI agents (such as GPT-5 and Claude) have dramatically improved their ability to identify and exploit vulnerabilities in blockchain smart contracts . Simulations have shown that approximately $4.6 million in stolen funds can be recovered, and previously unknown zero-day vulnerabilities have also been discovered.
+ AI's secret correction
YouTube has sparked controversy by conducting an experiment using AI to fine-tune videos (such as skin retouching) without creators' consent . This raises concerns about the authenticity of content and creators' control.
+ Gap in university education
It has been argued that universities should teach students "AI readiness"—the ability to build and deploy real-world systems—beyond mere "AI literacy (theory) ." Employers are increasingly seeking practical implementation skills over theoretical ones.
References
1.
November 2025 Korean Startup Investment Briefing - THE VC
2.
AI Literacy vs Readiness in Universities
3.
The VC's Expiration Date in the AI Era
4.
AI Music, Can You Tell The Difference?
5.
AI防衛革命：シリコンバレーの異端児「アンドゥリルがが日本の防衛・部品産業と協業するシナリオ
6.
Charts of the Week: If you can't join 'em, beat 'em
7.
Clouded Judgment 12.5.25 - Monopoly Money
8.
EU plans five AI gigafactories with 100,000 high-performance AI chips
9.
Google gathers triple OpenAI's AI data through its search monopoly
10.
Google, Nvidia, and OpenAI – Stratechery by Ben Thompson
11.
I Was Wrong About AntiGravity
12.
I love AI. Why doesn't everyone?
13.
LLMs Navigate Knowledge, They Do Not Create
14.
Local tech scenes have changed
15.
OpenAI declares 'code red' as Google catches up in AI race
16.
OpenAI tests “Confessions” to uncover hidden AI misbehavior
17.
AI agents find $4.6M in blockchain smart contract exploits
18.
Synopsys and NVIDIA Double Down on Acceleration
19.
The Enforced Conformity
20.
What Happens After the Hype? Lessons from Mobile Internet's Long Road to Success
21.
Where Do the Children Play?
22.
With Both Apple & AI, Timing Remains Everything
23.
YouTube secretly tests AI video retouching without creators' consent
24.
Are you looking for a star image device?
25.
👀 Did OpenAI's $100 billion path just get narrower?
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'!
