The Closing of the Digital Frontier

Growing up with the internet, I always felt it was a place where anyone could explore and innovate without barriers. However, with the announcement of the Anthropic Mythos, I'm experiencing a sense of loss akin to the closing of the American frontier described by Frederick Jackson Turner in 1893. Turner believed that the availability of free land shaped America's unique character, fostering liberty and ambition. Today, the digital realm was our modern frontier, offering economic mobility and individual agency, but now it seems to be closing.

The internet once allowed a 16-year-old with no credentials to create and innovate without needing vast capital or connections. This freedom is threatened as powerful AI models become restricted to the wealthy and well-connected, echoing Rudolf Laine's idea that those with capital have an insurmountable advantage in the age of labor-replacing AI. George Hotz calls this neofeudalism, where a small elite holds a monopoly on intelligence, much like a permanent underclass.

The comparison to nuclear weapons is flawed because intelligence is a creative force, not just destructive. Countries will pursue AI aggressively, and without global alignment on risk reduction, the consequences could be severe. Anthropic's decision to limit Mythos to enterprise partners like Microsoft and Cisco raises safety concerns, especially if a security breach occurs. The privatization of state-scale capabilities without accountability is troubling.

API access, while not full ownership, offers a programmable surface that fosters innovation. Restricting it stifles creativity and prevents latent capabilities from being tested in the real world. The reaction to AI finding security vulnerabilities is overstated; security is an arms race, and open access could enhance security systems. However, researchers can't access these models for rigorous safety studies, relying instead on open-source models.

Intelligence should be accessible like electricity, with safety guardrails to prevent misuse. Organizations with government-level capabilities should act with transparency and accountability, providing due process for access decisions. The current model of extracting value from a population without meaningful consent and concentrating benefits within a small circle is reminiscent of historical patterns of exploitation.

While this may seem bleak, there is hope that this is just a phase in AI's evolution, akin to the mainframe era before personal computing. Open-source models are advancing rapidly, and hardware supply chains could scale to make intelligence widely accessible. However, we must be cautious not to stifle innovation and creativity in the digital realm, just as cities cut down trees to avoid potential lawsuits.