46 provocative theses from an anonymous researcher: AI will upend everything, including nuclear deterrence and human dominion
- Core Thesis: The production of intelligence by AI may still have room to grow by 4 to 10 orders of magnitude. Algorithmic progress will trigger technological leaps far beyond expectations, reshaping society, the economy, and power structures in the short term, potentially leading to permanent class stratification and fundamental changes in national security paradigms.
- Key Elements:
- Algorithmic progress will bring about a "great leap." The production of intelligence may still have 4-10 orders of magnitude of improvement left, and current understanding underestimates the speed and depth of this transformation.
- AI self-improvement is entering a takeoff phase, algorithmic research is accelerating, and there is a "correction escape velocity." The performance of long-horizon task agents will overcome the bottleneck of error accumulation.
- The robotics field will experience its own "ChatGPT moment," but the scaling of physical robots globally (100 million units per year) may not be achieved until the 2030s.
- Automation will lead to profound deflation, with most goods approaching zero marginal cost. However, it may create a "permanent underclass," defined not by low income but by a lack of agency.
- Mutually Assured Destruction (MAD) may become ineffective in the face of AI and automated military supply chains. Military, police, and other government enforcement mechanisms will be automated and become much smarter.
- AI labs may face pressure to be nationalized, in order to reconcile the dangerous tension between state power and the private companies that control them, especially after they come to possess decisive tools of violence.
- The tech tree harbors genuine dangers (such as zero-day vulnerabilities and the fragile world hypothesis). Failures in future power accumulation and coordination could lead to tyranny or catastrophe.
Original Author: bayeslord
Original Translation: TechFlow
Introduction: bayeslord (@bayeslord) is an anonymous but influential figure in the AI × crypto space. He doesn't shill or chase trends, instead focusing on deep-diving into technical cores like scaling laws and algorithmic depth.
This blogger recently wrote a 46-point checklist projecting the future of technology, AI, and related fields. It argues that everyone is using past efficiency curves to understand AI, but the real leap hasn't come yet, and there might be four to ten orders of magnitude left in intelligence production.
He moves from algorithmic acceleration to robotics, capital, a permanent underclass, and finally lands on the sharpest point: Mutually Assured Destruction (MAD) might fail, police and military could be automated, and AI labs might be nationalized.
The original post has nearly 1 million views. Although the views are extreme, each point is relatively self-consistent and worth a read for tech-savvy audiences.
This list is based on a thread I posted on June 4, with some edits and additions. Several people said the original was hard to read, so I've compiled this version.
Intelligence
1. Algorithmic progress will catch everyone off guard. The entire world—markets, governments, militaries, companies, individuals—is using the productivity and patterns of recent years to understand AI's impact and judge how things will roughly go. Even new labs that claim to believe in 'recursive self-improvement' think it's just the old pattern with an agent running in a loop. It's not. I suspect there are still many orders of magnitude left in intelligence production, perhaps up to ten, with four to seven being more likely. In principle, exceeding ten isn't impossible, but that would heavily bump against what I suspect physics truly allows. Unlikely, but not ruled out. If this judgment holds, then the real direction of things is not what it appears to be on the surface; a big leap is approaching. Anything happening along this path will make the world far stranger than almost anyone is pricing in.
2. We are in the early stages of takeoff. AI improving AI could ultimately become the most consequential step in history. This is not a guarantee, as we don't know how far we are from the physical and computational limits of intelligence, but I'm betting we are still far away (as mentioned, squeezing another 4 to 10 orders of magnitude of intelligence output per unit of compute seems possible).
3. Since we've entered the takeoff phase, algorithmic research is accelerating. Compute remains a scarce resource, but the opportunity cost of a researcher's time has dropped because you can directly deploy an agent to run any task, even random ones. It might just bring something back. All new ideas carry an 'optimization debt,' which can now be repaid using unsupervised token consumption. Vast research scaling law curves will be traversed one by one.
4. AI models will continue to improve, frontier models especially. The only real wall is physics. Models are becoming more autonomous, smarter, and consistently better. Mathematics and code are being conquered by scaled reinforcement learning; everything else is next. The distinction between 'verifiable' and 'non-verifiable' will slowly fade as a meaningful boundary. Going forward, automated AI research and AI learning will increasingly look like the same thing. Training a model well is fundamentally closely related to the model learning well on its own. Sample efficiency, creativity, and all other limitations will be resolved, then approach algorithmic optimality at arbitrary scales.
5. The idea that long-horizon agents require equally long training data is wrong because generalization exists across time. Long tasks aren't just stacked ‘length’. This relates to LeCun's fallacy about (1-e)^n error accumulation. What actually happens is error correction. Error correction occurs at multiple scales simultaneously, from the level of single token generation to every step within a long task. The reason METR's graph is trending upward is partly because agents are starting to reach the escape velocity of error correction.
6. An engineering-grade deep learning science is about to emerge. It will push us towards the algorithmic maturity of AI much faster than most expect—although, as noted, how far this path can go in principle remains unclear. For instance, a science of scale invariance would drastically increase the scale and returns of useful experiments, because one experiment on a single GPU could tell you how to scale to one hundred thousand.
7. Every area of technical human activity will have its own 'move 37' moment (the move against Lee Sedol by AlphaGo that transcended human intuition), and then quickly, 'move 37' itself will seem outdated. I mean all fields.
8. Compute will continue to improve. Today's best matrix multiplication machines are far from the physical limits of AI accelerators. Digital silicon has significant room for improvement. There are also many candidates for new substrates; their algorithmic debt will be squeezed to the limit by automation, but we don't yet know which is optimal for AI in terms of space, energy, time, manufacturability, and cost. Photonics and stochastic silicon are interesting candidates, but I also expect the singularity itself to be surprising.
9. How far a lab can lead depends partly on the returns to automation and scale, including the returns from deeper algorithmic depth. If the practice (and theory) of deep learning remains shallow forever, then long-term moats probably won't be primarily algorithmic, as secrets are relatively easy to discover. Eventually, distillation plus data plus time can catch up to compute scale, perhaps a bit slower. Currently, we seem partly in this state, but even if that's true, no one guarantees it will stay this way.
10. If deep learning becomes less shallow with scale, then each increment of automation and scale will buy you algorithmic secrets that others find increasingly harder to reach. We currently seem to partly fit this state too. Both scenarios end when marginal utility returns saturate with scale and research. We don't know where that point is. It could be 2 orders of magnitude from today, or 20. No one knows.
The Intelligence Supply Chain
11. For at least the next few years, compute will be a fiercely contested resource. But during this time, it will begin to commoditize, and we will look back at the 2020s as a time of scarcity. Scale is expanding and working; capital follows, turning the flywheel repeatedly. More matrix multiplication machines, more fabs, more energy are on the way. The bottleneck in intelligence production is temporary. Potential economic speed bumps are not counted in this.
12. The nature of the intelligence supply chain is changing. Currently, it is highly concentrated in labs. But labs are automating the core thing that makes them strong—researchers, and the discovery of algorithmic advantages. Once this process begins, assuming open source follows not too far behind, especially if labs don't lock down their AI researcher models, then the labs' advantage will shift towards easier access to capital, more compute, proprietary data, business relationships, and better products. This indeed depends on how the algorithmic depth issue mentioned earlier resolves, among other factors.
13. Distributed training will reduce the need for massive single-data-center builds, giving some advantage to non-hyperscaler firms. However, in the pure dimension of single largest training runs, it won't surpass the hyperscalers.
14. Automated AI experimentation will cause algorithmic secrets to be broadly discovered, as these secrets are inherently easier to distribute than full-scale training. How far this path can go is unclear, but I expect it to be quite far. As mentioned, the fundamental depth of deep learning remains unknown, and the upper bound of this judgment depends on that unknown.
15. Despite these forces seemingly favoring academia and open source, they might still shrink due to the cost and opportunity cost of compute. For example, is a GB300 more valuable serving GLM5.2 or Fable, or running non-frontier research in an academic lab, or building Mythos 2 inside Anthropic? The market will identify where demand is highest, and currently, that place is indeed in labs. This means open-source labs might become even more compute-hungry, even if they have money, unless they have locked in compute capacity in advance. Even if they have, they must consider the opportunity cost of doing their own research versus renting out their compute. See the Colossus and Anthropic partnership for reference.
16. In an environment where AI capabilities start getting interesting (within the next 0 to 18 months), open source might also become socially difficult, especially if we are slow to accelerate safety—which we have been so far.
17. As capital flows towards labs, open source might start to shrink. There is a coordination problem here: aside from labs (and perhaps governments), no one wants a token monopolist. But if this problem can be solved and the regulatory environment is friendly, perhaps the outcome will be okay.
Robotics
18. Robotics will have its 'November 2022 ChatGPT moment', and then another moment like 'November 2025 Opus 4.5'. Neither has happened yet, but they are coming, and faster than people think, as a result of rapid AI progress, including AI-accelerated physical systems engineering. The gap between these two moments in robotics doesn't seem likely to be three years.
19. However, truly scaling up the physical quantity of robots worldwide might need to wait until 2030 or later. Although we build about 100 million cars per year, and humanoid robots are much smaller. Considering we also build 1 billion smartphones per year, if capital and algorithms run fast enough, achieving an annual production scale of 100 million robots by 2030 is reasonable. 10 million per year is definitely achievable; we are already doing it in the drone market. As long as software can prove at small scale that humanoid robots are worth the investment, it will unlock infinite capital, proportional to the quality of the proof.
20. What looks like hard ceilings for robotics today will disappear, including poor sample efficiency, relative data scarcity, expensive and difficult hardware design for hands and motors, the fractal complexity of the physical world, and the undocumented tacit knowledge about how we work in the world (e.g., plumbing). World models seem useful, but exactly which thing is unimportant. Research scaling laws will be ground down until diminishing returns set in.
21. Global demand for robots is easily in the tens of billions, especially when considering all form factors. There is so much manual labor worth automating. The market will figure out how to do this, and people probably won't stand in the way.
Progress
22. Science is being automated and virtualized. This means much of the progress the world needs will come from automated labs and simulations. We don't know the full computational limits of virtualization, but robot-driven labs in fields like biology and material science will remove many bottlenecks, pushing the boundary of 'verified virtualization' along the way to improve sample efficiency and the net returns of 'making it real'. In virtually every domain, we will have some combination of neural models, explicit simulations, and real-world experiments, jointly boosting the returns per dollar and per hour in fields like biology and material science.
23. Progress laws are everywhere. In deep learning, they are called scaling laws. It's hard to judge when any S-curve will saturate, and whether another S-curve lies beyond the horizon. What needs to be understood here is that the engine of civilization's progress itself has a progress law. Our progress is likely saturating, like most natural processes, but we actually don't know where the saturation occurs. The maturity of technology and civilization could be near or far. We are at a historical juncture: first, we have barely invested any resources into progress, but this is changing rapidly; second, we are automating the very machine that directly produces more progress. We live in interesting times.
24. A future of scaling up or scaling out. From zero to one or from one to n. How much progress the universe allows us in breadth and depth is an open question. Breadth is easier to estimate, as it's roughly 'how many more calculations do the laws of physics allow us to perform from now on?'. And how 'deep' that calculation can be—in the broadest sense of the word—is unknown. In some versions of the future, the tech tree is incredibly deep, and the accessible computational universe is so rich that we will keep inventing and discovering until physics stops us, if it can. Other versions are flatter: we quickly fill up a shallower tech tree, achieve technological maturity relatively easily, then scale it out until we are satisfied or physics gets in the way.
Capital and Production
25. More capital plus more intelligence means a more intensified capitalism, meaning we rush towards market equilibrium faster. In the long run, this should naturally lead to deflation, to most important goods competing down to near-zero marginal cost, including AI, food, housing, medicine, electronics, entertainment, and travel. Provided we don't let people stand in the way. In some cases, they probably will.
26. Mining will be automated. Transportation by land, sea, and air will be automated. Factories will be automated. Workers will be automated. Distribution centers will be automated. The maintenance, improvement, and expansion of the entire supply chain will be automated.
27. Some humans will retain jobs, for a very, very long time. What percentage of humanity this represents is an open question. Those who say this number will be high are overconfident, and those who say it will be zero are equally so. However, it's hard to imagine how long humans can continue to contribute marginally in the 'knowledge' part of knowledge work. Demand for some roles, like doctors, might drop dramatically—if we have a $20/month superhuman AI doctor, plus on-demand diagnostics, plus massive health improvements from medical technological progress. But because we currently cartelize doctors, we might continue to do so, making being a doctor a good job. Demand for entertainment will probably rise, but production costs will fall; the technical need for humans in entertainment is already greatly reduced. However, we care a lot about other humans, so perhaps we will continue to care, making being an actor more profitable. One way to think about how this evolves is to consider how many intermediary layers exist in the supply chain between today's worker and consumer. For a TikTok influencer, zero layers. For a doctor, zero layers. For a factory worker, many layers. Whether a job can be disintermediated, competed away, or substituted will likely determine its fate to a large extent. This analysis is quite nuanced, and this paragraph doesn't do it justice, but this all presumes we don't hit a cliff on the demand side—which could happen if too many people don't work, and productivity or government efficiency can't sustain UBI or universal basic healthcare.
28. Related to the above but not contradictory: a 'permanent underclass' might genuinely exist. In the better worlds where this happens, it looks more like severely constrained agency rather than devastated income. For most people, this is ultimately acceptable; our agency is already heavily constrained by modern society. But it requires psychological adaptation, which might take time and could be painful.
Culture and Psychology
29. The human mind currently grows and adapts slowly, but this will change. The key is for it to change for the better, though this won't be easy for some. Abundant intelligence and automation will allow us to engineer psychological structures far more durable than today's—which are evolutionary remnants maladapted to our environment. Psychiatry and psychology will undergo a millennium's worth of innovation within a few decades. Humans will fundamentally get better. Crude, degenerate 'pleasure direct connection' is overrated as a risk, because we will have more sophisticated and diverse mental engineering tools available.
30. In a world of extreme uncertainty, people will fight for power, status, and wealth more fiercely than ever, betraying their kind with a clear conscience. They will invent all sorts of reasons to explain why their actions are good, even great. Look around.
31. You will live to see embarrassment you cannot believe.
