Why Nobody Wants Kids Anymore?

NCM: The LLM Killer? Stanford May Have Changed AI Forever Hello world. I’m an unemployed ex-Big Tech software engineer with 25 years of experience in the tech industry. And one of the perks of involuntary early retirement is that I have plenty of time to follow the latest developments in AI. Which is great, because the AI industry seems determined to change the world every five minutes. But here’s something I’ve been thinking about for a while. Almost the entire AI conversation revolves around Large Language Models, or LLMs. Every few months, we get a bigger model, a more expensive model, a new benchmark, and another announcement about how this one is definitely going to change everything. Meanwhile, the companies building these models are pouring staggering amounts of money into GPUs, data centers, electricity, and cooling infrastructure. And yet, today’s LLMs have some fundamental limitations. What if the next major AI breakthrough doesn’t come from making LLMs bigger? What if it comes from building AI differently? Well, researchers at Stanford University have been exploring a new architecture called Next-Concept Modeling, or NCM, and the early results are interesting enough to deserve our attention. Could this be the beginning of the end for LLMs? Maybe. Maybe not. But let’s talk about why this research caught my attention. 1. The Problem With Today’s LLMs First, let’s talk about how conventional LLMs work. At their core, these models predict the next token based on the prompt. A token might be a word, part of a word, or another small unit of text. The model predicts a token, adds it to the sequence, and then predicts the next one. This process continues until the model has generated its response. In grossly simplified terms, it’s a huge autocomplete. But their autoregressive generation process has a major drawback. The tokens in an answer are generated sequentially. Suppose you ask an LLM to generate a 1,024-token response. The model generally needs to perform a forward pass for each newly generated token, meaning roughly 1,024 sequential generation steps. There are optimizations, such as caching previously computed attention states, speculative decoding, and other inference techniques. So this doesn’t mean the entire model is recomputed from scratch every time. Nevertheless, generating long responses requires repeated computation, and model weights and associated data must be accessed as needed throughout inference. With models containing trillions of parameters, that can get expensive very quickly. And if you enable a reasoning mode that generates additional intermediate tokens, that only increases the number of passes and the amount of computation required before you get your final answer. The result is an AI industry that increasingly depends on enormous amounts of computing power, memory bandwidth, and energy. Now compare that with the human brain. The brain performs an astonishing range of cognitive tasks while consuming roughly 20 watts of power. Today’s frontier AI systems can require vastly more energy and computing infrastructure, particularly when training and operating at scale. That comparison isn’t perfectly apples to apples. Biological brains and digital neural networks work very differently, and their capabilities aren’t directly comparable using a single energy metric. Still, the contrast raises an important question. Are we approaching intelligence in the most efficient way possible? I’ve argued before that if we see major algorithmic breakthroughs that move us closer to more capable reasoning systems, they might come from somewhere outside the architectures dominating today’s AI industry. And this is where NCM gets interesting. 2. Meet NCM: A Different Way to Generate Text Next-Concept Modeling takes inspiration from an idea used in another area of AI: diffusion models. If you’ve ever watched an AI image generator create an image, you’ve probably seen how it starts with noise or a rough representation an

Let me tell you about a development that should make anyone working in the software industry a little nervous. Imagine spending decades building a software company worth billions of dollars. You hire thousands of engineers, invest enormous amounts of money in research and development, and build a portfolio of applications used by millions of people. Then one guy comes along, fires up an AI coding agent, and starts cloning your products. Welcome to the future of software. One Guy vs. Adobe and Microsoft Recently, a developer named Brandon Thomas attracted attention with ArtCraft, a project aimed at recreating the functionality of popular Adobe Creative Cloud applications using AI-assisted reverse engineering. The project includes applications inspired by products such as Photoshop, Illustrator, and Premiere Pro. Related projects extend the same idea to familiar office applications, including Word, Excel, and PowerPoint. These are not polished, feature-for-feature replacements for every original application. But that’s almost beside the point. The important development is that AI is making it dramatically easier to reproduce software functionality without rebuilding everything from scratch through traditional software development. I tried PhotoCraft myself. It wasn’t nearly as polished as Photoshop, but it worked for many of my basic image-editing needs. I also experimented with AI-assisted reverse engineering on a simple chess application. The resulting clone came surprisingly close to the original. And here’s the part that really caught my attention: I didn’t have to manually analyze every feature and write the replacement code myself. I could give an AI coding agent the task and let it do much of the heavy lifting. As a former Big Tech software engineer and architect, I find this fascinating. And potentially terrifying. For years, I’ve argued that AI is making software less valuable. These developments suggest that the process could accelerate much faster than many people expect. How Can AI Reverse-Engineer Software Without the Source Code? To understand why this matters, we need to look at how software reverse engineering works. When developers build an application, they write source code using programming languages such as C++, Java, or Rust. That code is typically compiled into machine code, which contains instructions a computer can execute. The resulting executable doesn’t normally include the original source code in its original, readable form. Traditionally, reverse engineering software required specialized expertise. Engineers used tools to inspect executable files, analyze program behavior, recover information about functions and data structures, and reconstruct how different features worked. This was difficult, time-consuming, and expensive. AI coding agents are changing that equation. An AI agent can work through an application feature by feature, using specialized reverse-engineering tools to gather information about the program’s behavior and internal structure. Tools such as Ghidra and IDA can help analyze compiled binaries. An AI agent can use the resulting information to reason about how a feature works, implement a replacement, test the result, and repeat the process. The basic workflow looks something like this: * Analyze: Inspect the original application and identify its features. * Investigate: Use reverse-engineering tools to examine relevant parts of the executable. * Reconstruct: Infer how a feature behaves from the available evidence. * Reimplement: Write new code that reproduces the feature’s behavior. * Test and repeat: Compare the replacement with the original and work through the remaining gaps. Repeat this process across enough features, and you may end up with a functional alternative to the original application. There are important limitations. Compiled binaries don’t necessarily reveal everything about the original source code, and some information may be missing or difficult to recover. Comple

There’s something deeply unsettling about watching a machine accomplish something extraordinary and realizing that the accomplishment itself might be the beginning of a much bigger problem. OpenAI has been pushing the boundaries of AI-assisted mathematical reasoning, including producing formal proofs for difficult mathematical problems. These developments are impressive, but they’ve also raised uncomfortable questions about the future of mathematics and the people who practice it. Because beneath all the excitement about AI solving difficult problems, there’s a much bigger question we should be asking: What happens when AI becomes so good at thinking that humans stop learning how to think for themselves? I call this the Mathapocalypse. I’m not suggesting mathematics is about to disappear. Rather, I think what’s happening in mathematics offers us a glimpse of a potentially disturbing future for human intelligence. Let me explain. Mathematics Is About More Than Getting the Right Answer I have a soft spot for mathematics. Back in college, I came pretty close to majoring in it. To me, math is a language that describes reality. There’s something elegant and beautiful about the simple truths humans have discovered through it. Mathematics is also a great example of a field where the process of discovery can be just as valuable as the final result. In many professions, the primary goal is to produce an output. A software engineer writes code. An architect designs a building. An accountant prepares a financial statement. The output is what the customer or employer generally pays for. If an AI can produce the same result faster, cheaper, and with fewer mistakes, it becomes difficult to justify paying a human to do the same work. But intellectual pursuits can be different. When mathematicians work on a difficult problem, they aren’t simply trying to produce a proof. They’re exploring ideas, developing techniques, testing conjectures, and discovering relationships they didn’t previously understand. Even when a particular approach fails, the effort can produce knowledge that becomes useful somewhere else. Sometimes, the real value lies not in the destination, but in the journey. Some of Humanity’s Greatest Discoveries Were Unexpected Side Effects Think about the history of mathematics and science. The development of calculus gave humanity powerful new ways to describe motion, change, and physical systems. Those ideas eventually became essential to engineering and physics. Number theory, once viewed by many as an abstract pursuit with limited practical applications, eventually became fundamental to modern cryptography. Fourier analysis, developed to study mathematical problems involving heat, became an important tool in signal processing, communications, and image compression. These developments didn’t happen because someone sat down and said, “Hey, let’s invent the technology that will power smartphones in a couple of centuries.” They happened because people pursued difficult intellectual problems, sometimes for reasons that seemed entirely disconnected from practical applications. The journey produced tools that nobody could fully anticipate at the beginning. Consider the Space Race. The competition to reach space generated advances in rocketry, satellite communications, integrated circuits, and other technologies that eventually found applications far beyond the original mission. Humanity didn’t just get astronauts into space. It developed capabilities that transformed life back on Earth. When humans investigate difficult problems, we sometimes discover things far more valuable than the answers we were originally looking for. Now imagine replacing that process with an AI that simply generates the answer. Would we still get the same unexpected discoveries? The answer is obviously NO. But there’s an important distinction between a machine producing a solution and humans understanding the process that led to it. And that distinction could b

I’ve spent the last few days obsessing over a Huawei smartphone. Before you worry that early retirement has turned me into some kind of consumer electronics slop fanboy, let me explain. I don’t particularly care about the phone itself. What interests me is the chip inside it, the Kirin 9050 Pro, and a manufacturing technique that could have enormous implications for the global AI race. It’s called Tau Scaling, and if the technology works as described, it could undermine one of America’s biggest advantages over China: access to advanced computing hardware. But here’s where things get really interesting. I think the AI race could end with neither America nor China achieving a decisive victory. Instead, both countries could spend increasingly absurd amounts of money chasing smaller and smaller improvements until the economics stop making sense. In other words, we might spend hundreds of billions of dollars trying to build the world’s most powerful AI, only to discover that nobody can afford to stay ahead. Welcome to the future. America’s Secret Weapon: Computer Chips Let’s start with the basics. Artificial intelligence depends on three fundamental ingredients: * Algorithms: The techniques used to build and train AI models. * Data: The information used to teach those models. * Compute: The computing power required to train and run them. America and China are competing across all three dimensions. But for a while, America has had a particularly important advantage in compute. Training powerful AI models requires enormous numbers of calculations. Modern AI accelerators, such as Nvidia’s GPUs, are designed to perform these calculations efficiently and at scale. And the more computing power you can bring to bear on training, the more capable your models may become, all else being equal. This is why access to advanced semiconductors matters so much. One way to increase a chip’s computing potential is to pack more transistors into a given area. Modern manufacturing techniques allow chipmakers to fit billions of tiny electronic components onto a piece of silicon. The industry often describes these manufacturing technologies using nanometer process nodes. Although those labels are not literal measurements of every transistor or feature, newer process generations generally enable improvements in transistor density, power consumption, and performance. Companies like TSMC have pushed semiconductor manufacturing to extraordinary levels. And America has tried to use its advantage in advanced chips to slow China’s progress in AI. The Plan to Keep China Behind The American strategy has included restrictions on exports of advanced Nvidia AI chips to China, along with restrictions on semiconductor manufacturing equipment. One particularly important technology is extreme ultraviolet lithography, or EUV. EUV machines use extremely short-wavelength light to print incredibly fine circuit patterns onto silicon wafers. They are among the most sophisticated machines ever built, and ASML, a Dutch company, is the sole commercial supplier of EUV lithography systems. Without access to these machines, producing cutting-edge chips becomes significantly more difficult. China has been working with older deep ultraviolet lithography, or DUV, equipment, while pursuing increasingly sophisticated manufacturing techniques of its own. The restrictions have made China’s path harder and more expensive. But they have not stopped China from trying to close the gap. Chinese companies have explored ways to connect large numbers of less powerful chips, obtain computing capacity from overseas data centers, and smuggling older Nvidia AI chips into the country. Imagine two competitors in a race. One has access to the latest equipment, an enormous supply of high-performance chips, and the ability to build massive AI data centers. The other has to work with less advanced equipment and find clever ways to compensate for the difference. The obvious American strategy is to keep the

The other day, I was flying my drone when my dad called me, completely panicked. Someone had called my mom claiming to be from the Brooklyn Police Department. They told her she was somehow implicated in a money-laundering investigation in Shanghai. Then they tried to transfer her to a supposed Chinese law-enforcement officer who would “verify” her identity. I immediately yelled: “NAAAAAAH, Mom! Don’t say anything! Hang up! It’s a scam!” Thankfully, she did. But afterward, I couldn’t stop thinking about it. Why does everything feel like a scam now? Obviously, there are always going to be criminals trying to steal your money. But I increasingly wonder whether these scammers are actually a symptom of a much bigger problem. Because the feeling that you’re constantly being manipulated, extracted from, upsold, tracked, or squeezed doesn’t stop with scam phone calls. It seems to be everywhere. The Crapification Cycle Think about the products and services we use every day. A product starts out relatively simple and useful. Then something happens. The quality gets worse. The price goes up. Features that used to be included become subscriptions. Repair becomes harder. Companies introduce artificial limitations. And eventually, the thing you bought isn’t really a product anymore. It’s a recurring revenue stream. Your refrigerator needs a subscription. Your software needs a subscription. Your car increasingly wants a subscription. Your streaming services need subscriptions. Even products that were traditionally one-time purchases are increasingly being turned into services that require you to keep paying forever. And if the product is “free,” there’s a good chance you are the product. Your attention becomes the commodity. Your personal information becomes the commodity. Your behavior becomes the commodity. The entire system starts looking like: Lower quality → higher prices → more fees → more data extraction → more consumption. I call this the Crapification Cycle. But Why Does Everything Feel So Expensive? Then there’s the concentration of corporate power. Walk through a supermarket and you’ll see hundreds of brands. It looks like you’re surrounded by competition. But many of those brands are ultimately owned by a surprisingly small number of enormous corporations. That matters because competition is supposed to force companies to give consumers better products at better prices. But when fewer companies control larger portions of a market, that pressure becomes weaker. Companies can raise prices. Shrink the product. Cut customer service. Reduce quality. Add fees. And sometimes they can do all of this while still reporting enormous profits. You end up with the strange experience of paying more while receiving less. More money. Less product. Worse service. And then there’s something even bigger. The Financialization of Everything Increasingly, companies aren’t just selling you products. They’re selling you financing. Debt. Subscriptions. Insurance. Future revenue. Financial products. Housing isn’t just somewhere to live. It’s an investment vehicle. Healthcare isn’t just about keeping people healthy. It’s an enormous revenue-generating industry. Education can become a vehicle for decades of student debt. Even businesses that traditionally existed to sell physical products increasingly have enormous incentives to maximize financial metrics rather than simply make better products. Companies can buy back their own shares instead of investing in research. Executives can be rewarded for increasing stock prices. Short-term financial performance can become more important than building something durable. And eventually, the economy starts to resemble a giant casino. Money chases whatever asset, company, technology, or narrative promises the biggest future return. The underlying reality can become almost secondary. We’re no longer just consumers. We’re participants in a gigantic financial system that increasingly seems designed to extract value from