Two Technology Stories That Could Define the Next Decade
Every so often, the technology world produces a week that feels genuinely pivotal. This is one of those weeks. Two separate stories have broken through the noise to capture the attention of researchers, investors, and everyday technology enthusiasts alike. The first involves a bold claim from an AI startup that says it has cracked a mathematical problem holding back large language models for nearly a decade. The second involves a man living with ALS who has become what researchers are calling "the first power user" of a brain-computer interface implant. Taken together, these stories paint a picture of a technology landscape moving faster than many of us can fully process.
The LLM Bottleneck: A Problem Most People Never Knew Existed
To understand why the claim made by AI startup Subquadratic matters, you first need to understand the problem it says it has solved. Large language models — the engines powering tools like ChatGPT, Claude, and Google Gemini — are built on a technology called the transformer architecture. Transformers are extraordinarily powerful, but they carry an inherent mathematical inefficiency that has been understood and quietly tolerated by researchers for years.
The core issue is computational scaling. As a transformer processes longer sequences of text, the number of calculations it needs to perform does not grow in a neat, manageable straight line. Instead, it grows quadratically — meaning the workload balloons at an accelerating rate as input length increases. This has real-world consequences. It makes LLMs expensive to run, energy-hungry, and slower than they could otherwise be. It also places a practical ceiling on how long and complex a conversation or document an LLM can efficiently handle.
For years, this has been the quiet tax imposed on every AI application built on transformer technology. Startups and research labs have chipped away at the edges of the problem, but no one has claimed to have eliminated it at the architectural level — until now.
What Subquadratic Is Actually Claiming
Subquadratic, which recently emerged from stealth mode, says it has developed a new approach that dramatically slashes the number of computations transformers need to carry out when generating responses. According to the company, the result is a large language model that is faster, significantly cheaper to operate, and consumes far less energy than any comparable model currently available on the market.
These are extraordinary claims, and the AI research community has responded with a predictable mixture of curiosity and skepticism. Reducing computational complexity at this level is not the kind of thing that happens quietly or without controversy. If the claims hold up under scrutiny, the implications are profound — not just for the companies building AI products, but for the environmental conversation around AI's growing energy footprint, which has become an increasingly urgent concern.
To its credit, Subquadratic has begun sharing early evidence to back up its assertions. Researchers who have examined this preliminary data suggest the approach may be genuinely worth paying attention to, even if full independent verification has not yet taken place. The scientific process of scrutiny is ongoing, and the community's skepticism is healthy and appropriate. Nevertheless, the early signs are interesting enough that dismissing the claim outright would be premature.
Why This Matters Beyond the Lab
If Subquadratic's approach proves out, the downstream effects could be significant across multiple industries. Consider the following:
- Businesses running AI-powered customer service, code generation, or content tools could see their operating costs drop substantially, making AI adoption more accessible to smaller companies.
- Developers building applications on top of LLMs could work with much longer context windows without hitting performance or cost ceilings.
- The environmental case against large-scale AI deployment, which centers on energy consumption in data centers, could become considerably weaker if models genuinely require a fraction of their current computational load.
- Edge deployment — running AI models directly on devices rather than in the cloud — could become viable for a far wider range of applications.
None of these outcomes are guaranteed. But the conversation Subquadratic has started is one the industry needed to have.
Brain-Computer Interfaces Enter a New Era
Shifting from software to hardware — and from the abstract to the deeply human — the second major story of the week centers on brain-computer interface technology and a man named Casey Harrell. Harrell has ALS, a progressive and devastating neurological disease that gradually robs people of the ability to move and communicate. He has become what researchers are describing as the first true "power user" of a brain implant.
Brain-computer interfaces, or BCIs, work by detecting electrical signals produced by neurons in the brain and translating them into digital commands. The technology has been in development for decades, but clinical trials have historically been limited in scope, duration, and the complexity of tasks participants could accomplish. What is changing now is the pace and ambition of those trials.
Why BCI Trials Are Accelerating in 2026
Several converging factors are driving the current surge in BCI research and clinical activity. Improvements in miniaturization mean that implants are becoming smaller, safer, and less invasive. Advances in machine learning have made it significantly easier to decode the complex patterns of neural activity into meaningful, reliable commands. And a growing number of companies — motivated by both genuine humanitarian goals and commercial opportunity — are investing in the space at a level that simply was not possible five years ago.
Casey Harrell's story is compelling not just because of the technology itself, but because of what it represents for people living with conditions like ALS. The ability to communicate, to control digital environments, and to maintain a degree of independence and agency is not a luxury for these individuals — it is transformative in the most literal sense of the word. Trials that demonstrate sustained, real-world utility for patients are the proof of concept the field has needed to attract broader support and regulatory confidence.
Looking Ahead: Convergence and Caution
What makes this particular moment in technology so striking is the way these two stories — one about the efficiency of AI models, the other about direct interfaces between human brains and machines — point toward the same horizon. As LLMs become cheaper and less energy-intensive, deploying sophisticated AI processing at the edge becomes more practical. As BCI hardware becomes more capable and more widely trialed, the idea of AI systems that work in close, real-time partnership with human neural activity moves from science fiction toward clinical reality.
None of this means the challenges are small. Subquadratic still needs to convince a skeptical research community, and BCI technology still faces formidable regulatory, ethical, and technical hurdles before it reaches anything like mainstream adoption. But the direction of travel is clear, and the pace is accelerating. For anyone paying attention to where technology is headed, this is a week worth remembering.