The Riemann Hypothesis Solved by AI, Peer Review Standardized, and LLMs Begin Self-Replication—What Happens to SMEs Amid the ‘Collapse of Intellectual Labor Costs’
Related Articles
The Day When the 3 Million Yen Paid to Experts Becomes 300,000 Yen
Consider three pieces of news:
- AI assisted in proving the Riemann Hypothesis. Levent Alpuoglu, a mathematician at Anthropic, used Claude to make breakthroughs on this extremely difficult mathematical problem that had remained unsolved for 167 years in about six weeks.
- Academic peer review is being standardized by AI. The overlap rate of peer reviews for ICLR jumped from 12.0% in 2026 to 21.3% in 2027. The ‘eyes’ of reviewers are beginning to be replaced by AI.
- Research has been published on LLMs self-replicating. The models output their own ‘genes’ to train the next generation of models. This structure allows AI to nurture AI without human intervention.
These three pieces of news may seem unrelated, but they point to the same conclusion.
“The cost of intellectual labor is structurally collapsing.”
This is not just a story about large corporations. It is the most significant change that small and medium-sized enterprises (SMEs) should confront head-on right now.
—
The ‘Real Shock’ of the Riemann Hypothesis Being Solved
The mathematical significance of the Riemann Hypothesis is honestly irrelevant to most business leaders. The issue is not there.
What is noteworthy is the time frame of “six weeks.”
For 167 years, the world’s genius mathematicians could not solve this problem, but one mathematician, in collaboration with AI, broke through in just six weeks. Of course, it is a given that Alpuoglu is an exceptional mathematician. However, he himself acknowledges that he could not have reached this point without AI.
The structure to be gleaned from this is as follows:
The productivity of “super-advanced expertise × AI” has skyrocketed to tens or hundreds of times that of conventional methods.
If you think this is happening only in the world of mathematics, you are mistaken. The same phenomenon is beginning to occur in law, accounting, design, marketing analysis—every field where money has been paid for the brains of experts.
To be specific:
- Legal checks on contracts. Hiring a lawyer costs between 50,000 and 150,000 yen per case. An AI draft review can be done for just a few hundred yen, and a final check by a lawyer would cost 20,000 to 30,000 yen.
- Preparing subsidy application documents. If you outsource to a consultant, it costs between 300,000 and 800,000 yen. By using AI to create an outline and finishing it in-house, outsourcing costs can be brought close to zero.
- Financial analysis reports. The monthly fee of 50,000 yen paid to an accounting firm is already being replaced by a world where reports are automatically generated using accounting software and AI.
What happens when the cost of “outsourcing to experts” becomes one-tenth? This is what business leaders need to consider.
—
The Standardization of Peer Review Indicates the ‘Commoditization of Expert Judgment’
At first glance, the discussion about academic peer review may seem like an event occurring within universities. However, when you extract the essence, it provides insights that SMEs cannot afford to ignore.
The increase in the overlap rate of peer reviews from 12.0% to 21.3% at ICLR signifies what?
It means that “expert judgment” is being standardized and commoditized by AI.
As reviewers begin to use AI, similar comments and evaluations are being mass-produced. In other words, the value of the “professional eye” is starting to decline.
Let’s translate this into the context of SMEs.
For instance, in website design reviews. Previously, businesses paid between 100,000 and 300,000 yen to have their work evaluated by designers or consultants with a “professional eye.” However, if AI can automatically point out UI/UX issues, the “professional eye” can be replaced by a tool costing a few thousand yen per month.
Quality control inspection reports, evaluations of marketing strategies, and assessment sheets for hiring interviews—any task that can be patterned will be absorbed by AI.
What’s important here is that this is not about AI perfectly replacing humans. What the standardization of peer review indicates is the fact that AI can now generate a large volume of “80-point judgments” at almost zero cost.
For tasks where an 80-point judgment is sufficient, there is no longer a reason to rely on experts.
So, what should be entrusted to experts? The remaining 20 points—namely, “truly difficult judgments,” “the ability to read context,” and “the readiness to take responsibility.” This is where money will be spent in the future.
The task for SME leaders is clear. They need to take stock of their outsourcing partners and categorize tasks into “those that can be handled at 80 points” and “those that require 100 points.” The former should be switched to AI, while the latter should focus on selecting even better experts for concentrated investment.
—
Self-Replicating LLMs—’Zero Marginal Cost of Intellectual Labor’ Is Not a Fantasy
LLMs self-replicating may sound like science fiction, but the research behind it is much more grounded.
Models compress and output their own knowledge structures, using them as “seeds” to train the next model. This means that AI is beginning to take over parts of the traditional pipeline where humans collect data, label it, and train.
What lies ahead when this becomes practical?
The development costs of AI itself will decrease.
Currently, training a high-performance LLM from scratch costs billions to tens of billions of yen. However, once the cycle of self-replication and self-improvement begins, the cost of “producing” models specialized for specific applications will drop dramatically.
What does this mean for SMEs?
It means that the era of acquiring “company-specific AI” at a realistic investment level is approaching.
Right now, creating a business-specific AI model can cost several million to tens of millions of yen. However, as self-replication technology advances, the cost of generating “child models” based on generic models and company data could drop to tens of thousands to hundreds of thousands of yen.
Imagine this:
- An “estimation AI” that has learned from your company’s past ten years of estimates produces rough estimates for new projects in three seconds.
- A “field response AI” that has learned your entire construction manual answers questions from new employees 24/7.
- A “CS response AI” that has learned from your customer interaction logs automatically handles 70% of inquiries.
All of this could operate in a world where it costs only a few tens of thousands of yen per month. Systems that large corporations used to build for tens of millions of yen could now be accessible to SMEs.
This is where the “structure that allows SMEs to win” lies.
Larger corporations face costs of deliberation, adjustment, and company-wide deployment due to their size when implementing AI. In contrast, SMEs can act as soon as the president says, “Let’s do it.” When the speed of decision-making combines with the democratization of technology, the disadvantages of scale can be overturned.
—
So, What Should We Do?
To summarize the discussion so far, three things are happening:
- The production costs of specialized knowledge are collapsing (the case of the Riemann Hypothesis).
- Expert judgment is being commoditized (the case of peer review standardization).
- The reproduction costs of AI itself are beginning to decrease (the case of self-replicating LLMs).
When these three trends progress simultaneously, the market for “costs associated with intellectual labor” will fundamentally change.
What SME leaders should do now is threefold:
① Take stock of outsourcing costs. Professionals, consultants, production companies—identify which tasks among your current outsourcing costs can achieve 80 points with AI. Start by switching one task next month.
② Organize your “company data.” Estimates, manuals, customer interaction logs, past proposals—these will fuel your future “company-specific AI.” Keep them organized and accumulate them digitally. This is something you can start doing today.
③ Experiment on a small scale. Introduce one AI tool costing 20,000 yen per month and test it on one task. If it works well, expand its use; if not, discontinue it. There’s no need to establish an “AI strategy” like a large corporation. Just experiment and make decisions based on the numbers. That’s all you need to do.
—
What ‘Zero Cost of Intellectual Labor’ Will Destroy and Create
Finally, let’s talk about the future.
As the cost of intellectual labor approaches zero, the value of “possessing knowledge” will decrease. Instead, what will rise in value are the areas of “what to ask,” “what to judge,” and “who to build trust with,” which are still challenging for AI.
What will be destroyed are business models that thrived on “information asymmetry.” The structures that profited from hoarding specialized knowledge and receiving high rewards will collapse the moment AI democratizes knowledge.
What will emerge are players who compete on “the speed of judgment and execution.” Knowledge will be accessible to everyone. The differentiation will come from how that knowledge is used and how quickly actions are taken.
SMEs have always had an advantage in “the speed of judgment and execution” over large corporations. Now that AI is becoming an affordable weapon, this advantage will grow even larger.
The question is simple.
Will your company ride this change? Or will it find itself on the side of those who have “ended up on their own” when they finally realize it?
You don’t need six months to find the answer. Start by trying out one AI tool next week.
JA
EN