ComputePilot @ComputePilot

The conflict in Iran just proved one thing: Data centers are the modern castles. As they transition into critical infrastructure, they are becoming the primary targets in geopolitical warfare.

The value of engineers is shifting from writing code to designing how systems actually run.

Today, AI is mostly built and consumed in centralized data centers. Tokens are generated in one place and used through cloud APIs. For AI to truly enter everyday life, token use likely needs to become distributed，running quietly on our own devices, not just as a remote service we call.

General-purpose technologies usually follow the same path: first, gains are concentrated; then they spread; finally, costs fall and real productivity takes off. The electricity analogy is perfect. Early on, only power companies and large factories benefited. Building plants and grids required huge capital, and ordinary households couldn’t afford it. As grids expanded, homes got electric lights and power became basic infrastructure. When electricity prices fell, appliances like refrigerators and washing machines spread widely, and consumer society changed. AI today looks more like the early infrastructure phase: massive data center buildout, but not yet at the point where service costs drop enough to unleash a wave of truly AI-native products.

A big part of programming used to be translation: turning product ideas into code. Now LLMs are excellent translators. If your value is just writing runnable syntax, you’re competing with software that never sleeps. The edge shifts to understanding why something should be built and how the whole system works. The future “programmer” looks less like a coder, more like a systems designer with execution power.

Real productivity leaps don’t come from “doing the same work with fewer people.” History shows us 3 patterns: 1.Systems run continuously Factories didn't just swap steam for motors; they redesigned for 24/7 flow. 2.New demand becomes affordable Shipping containers made transport so cheap they unlocked global trade. 3.Experimentation gets cheap Software made A/B testing instant, killing the "guesswork" of the past. If AI is only used to cut headcount, we’re still in the steam age. The real winners are those using AI to do what was previously impossible, too slow, or too expensive.

In the AI era, architectural inertia shows up in three ways: 1.Decisions stay with humans, so AI can only advise. 2.Data is owned by departments, not owned by system. So there is no continuous learning loop forms. 3.Old processes are “AI-augmented,” not AI native redesigned. They are making the old system stronger and real AI-native models impossible.

For most companies, the biggest blocker to AI at scale is still data fragmentation. Without a stable, unified system that automatically generates and shares data, AI stays stuck as an analysis tool instead of becoming part of how the business runs.

Clearly another example that Robotaxi doesn’t scale like a network product. It scales like infrastructure. It is constrained by regulation and social trust, not just technology readiness.

Sawyer Merritt @SawyerMerritt

8 months ago

NEWS: New York Governor Kathy Hochul has pulled a proposal that would have allowed for commercial robotaxi services outside New York City. The proposal, which Hochul had included in her budget proposal last month, would have allowed autonomous-vehicle companies such as Waymo to

145 61 708 128K 45

AI needs a continuously running system. Most enterprises still operate in human-broken system silos. That gap is why AI struggles to scale.

AI at scale is no longer limited by model capability. The real bottleneck is organizational: whether companies are willing and structured to let systems run decisions, not just generate insights. Technology enables it but institutions determine if it scales.

The AI valuation debate isn’t about technical capability. It’s about whether and how fast traditional firms move from using AI as a tool to embedding it into decision systems. The real upside depends on rewriting organizational structures, and that usually requires competitive pressure to force the shift.

Big tech leaps don’t come from plugging new tech into old workflows. Early factories used electricity like “steam with wires” and saw limited gains. The real productivity jump came when they redesigned plants, processes, and decision-making around electric motors. Same pattern today.

AI’s real revolution isn’t smarter tools. It’s building decision systems where humans set goals, systems run the day-to-day choices, and feedback keeps improving them. Without that, AI is just a better tool, not a productivity shift.

Standard software is no longer where differentiation lives. Differentiation moves to AI-driven, data-specific decision layers.

Rohan Paul @rohanpaul_ai

8 months ago

Mark Cuban on the next job wave. Customized AI integration for small to mid-sized companies. "Software is dead because everything's gonna be customized to your unique utilization. Who's gonna do it for them... And there are 33 mn companies in the US."

260 715 8K 6.6M 8K

Seedance-type tools don’t fail on technology, but on commercialization boundaries. Mass, unfiltered generation runs straight into copyright and style disputes. The sustainable path is B2B: custom, IP-safe tools for creative companies. Character generation is high-risk; motion generation is safer, more scalable, and delivers real efficiency.

Those are robotic-assisted surgical systems only, without autonomy.

Rohan Paul @rohanpaul_ai

9 months ago

🇨🇳 In China, doctors sat in Shanghai and carried out surgery in Kashgar, nearly 5,000km away, using Robots. They used China-made 5G surgical robot, finishing the operation in 1 hour. With minimal blood loss and no complications.

30 83 572 80K 71

New art project. Train and inference GPT in 243 lines of pure, dependency-free Python. This is the *full* algorithmic content of what is needed. Everything else is just for efficiency. I cannot simplify this any further. gist.github.com/karpathy/8627f…

AI reshapes corporations by shifting value inside the operating process. Value moves away from executing work and toward designing the systems that execute it. Pickers give way to those who design and run warehouse systems. Manual tuning is replaced by automated process control in manufacturing. In sales, deal engines replace manual follow-ups. Across all of them, value shifts from execution to system design.

World models are built from consequences. To learn consequences, you need to take risks. But to be allowed to take risks, you already need to understand consequences. It is a paradox.