In a tweet a while ago, Mr. Karpathy shared his optimism for how AI can democratize access to knowledge and bring power to the people. Historically, major technological shifts rarely benefit individuals to the same degree as large organizations or the wealthy. Yet LLMs bring expertise cheaply and instantly. As he pointed out: even as a billionaire, you have access to the exact same model. Money cannot buy a better ChatGPT.
Mr. Karpathy is a tech figure I’ve admired for years, and I appreciate his enthusiasm and techno-optimism. In this ideal world, AI brings fire to humanity like Prometheus.
The reality, however, feels more nuanced.
I’d like to draw a parallel between AI and the Internet—the last major innovation that inspired hope for democratized access to the world. The vision was that everyone would have equal access to Google, Wikipedia, and interconnected digital libraries sharing all of humanity’s knowledge. I remember reading John Perry Barlow’s manifesto many years ago: “I come from Cyberspace, the new home of Mind.” I was excited and moved, believing the Internet would give us a truly free world.
Fast-forward to today, and we have witnessed power on the Internet concentrate even further. A few multinational conglomerates took control over most gateways to the digital world. Instead of remaining a free distributor of information, parts of the web transformed into the most efficient censorship and surveillance tools the world has ever seen.
History often rhymes, and AI is no exception: it is an amplifier, and it amplifies inequality too. Yes, young tech entrepreneurs can launch startups, tech-savvy individuals can build custom software, and corporations can automate their workflows. But as with every technological shift before, the future is not distributed evenly.
Price is already becoming a barrier: a $200/month plan is a steep requirement for serious knowledge work, especially outside the US.
While it might seem obvious in Silicon Valley, most of the world is still just getting to know AI. Even if LLMs are equally powerful on paper, not everyone has the means or background to harness them. Industry experts understand the nuance of their fields, have access to proprietary data, and can amplify its value through a data flywheel.
Furthermore, not everyone is equally equipped to leverage AI constructively. While tech elites see opportunities to build businesses around AI, many people lack the guidance to navigate such a powerful tool. Some rush to generate reports or assignments without verification, delegating their thinking entirely and eroding their critical faculties. In much the same way that short-form video and social media can impede focus, AI can amplify capability—or it can amplify passivity and distraction.
We are already seeing this dichotomy play out: students use AI to glide through coursework without deep engagement, while the software industry—anticipating workforce reductions—is scaling back entry-level hiring. The end result is that new computer science graduates who relied on AI shortcuts during school are now facing a significantly higher bar for entry into the industry.
We can paint a dystopian picture of the future where, like many technological innovations before it, AI further concentrates power among the few while leaving others behind. Gatekeepers dictate the terms of access, while AI replaces workers and causes broader dislocation, much like Citrini Research predicted. We could end up in a bizarre scenario where the S&P 500 reaches record highs while mass layoffs shrink the spending power of white-collar workers.
I am sure these are not the outcomes AI visionaries intended.
It doesn’t have to be this way. Alan Kay famously said, “The best way to predict the future is to invent it.” We still have the agency to shape what comes next.
This is why public awareness of AI—both its capability and its potential harms—is crucial. Disseminating the knowledge required to understand and effectively use AI is a necessary first step toward empowering people.
Open-weight models, local inference, and an open ecosystem matter now more than ever. They serve as essential alternatives that foster competition and prevent total enclosure. Open-source software helped shape the Internet, and it must continue to shape the evolution of AI. We shouldn’t submit to FUD or gatekeeping. An open and competitive AI research ecosystem is the surest way to bring power to the people—and we need it urgently.
The future is uncertain and the stakes are high, but I remain optimistic that we can turn Karpathy’s vision into reality. I don’t claim to have all the answers for how we get there, but the first step in solving any problem is recognizing that one exists.