If Timelines Are Short, Do Projects Now

Many people in the AI Safety community, and increasingly in the world at large, hold a belief that transformative AI could be developed on a timescale measured in years, not decades. We can argue about the specific probability distributions, but a significant portion of the mind-space concerned with AI x-risk operates on timelines that are uncomfortably short.

The logical conclusion is straightforward: If you believe we have little time, our actions must reflect extreme urgency.

Every month we spend waiting, debating, or navigating bureaucratic hurdles is a month we don’t get back. When your model of the world includes a non-trivial probability of game-over scenarios within the next decade, the value of time becomes astronomical. In such a world, the single greatest sin is inaction.

And yet, look at our field's structure. It is, for the most part, a structure built for peacetime, for long timelines. It is a structure of academic cycles, competitive fellowships, and grant applications. It is a structure of permission.

The Bottleneck is the System

The core problem is that we have many more motivated people than there are places where meaningful contributions can be made, and to get into these places, usually, some time or a lot of time must pass. This makes sense in a world with long timelines and a need to carefully allocate scarce resources like funding and senior mentorship. It makes far less sense in a world where the most scarce resource is time, and the primary untapped resource is a massive global pool of talent willing to work.

Consider the numbers. BlueDot plans to introduce 100,000 people to alignment fundamentals. The number of people aware of AI risk is in the millions. Yet the number of full-time researchers is in the hundreds. There is an army of potential contributors standing on the sidelines, not because they lack skill or motivation, but because the entry points are narrow, slow, and hyper-competitive.

We are turning away people who might be in the 90th percentile of research ability because they aren't in the 99th. We are sidelining someone who might be a 99th-percentile operations manager because our pipelines are almost exclusively designed for researchers.

If timelines are short, this is a catastrophic, self-imposed bottleneck. We are artificially constraining the surface area of our search for solutions.

Towards a Permissionless On-Ramp

If the problem is a permission-based bottleneck, the solution must be a permissionless on-ramp. We need infrastructure that allows motivated people to start contributing right now, without waiting for a fellowship acceptance letter or a funding decision.

This is one of the core theses behind an organization like Theomachia Labs. The goal is to build a platform that decouples the act of doing useful work from the act of being credentialed by an elite program.

The model is simple and radical:

1. Absorb the bycatch: Instead of filtering out 95% of applicants, create a structure where they can immediately contribute as volunteers. Passionate contributors have proven their ability to produce significant results in programs like the AI Safety Camp. Theomachia Labs aims to create a permanent home for this energy, turning a trickle of talent into a flood.

2. Coordinate, don’t scatter: Instead of dozens of disconnected fellowship projects running in parallel, operate under a unified, long-term research agenda guided by domain experts. This allows for sustained, multi-phase projects that can tackle harder, more complex problems than an 8-week sprint ever could. It replaces a fragmented, hits-based approach with a coordinated, strategic one.

3. Value the full stack: An effective response to AI risk requires more than just researchers. It needs engineers, communicators, project managers, and operations specialists. A permissionless structure can and must create valuable roles for all of them. Theomachia Labs is designed as an ecosystem, not just a research pipeline, leveraging talents that other programs overlook.

4. Build a launchpad: For many, the biggest hurdle is the experience gap. "I can't get a job without experience, and I can't get experience without a job." A volunteer-powered organization solves this. Contributors gain real project experience, a public portfolio of work, and a network of collaborators. They are no longer an unknown quantity in a pile of applications; they are a proven asset. This becomes their launchpad into full-time roles, either within the growing organization or elsewhere in the ecosystem.

This model is not a replacement for elite fellowships, which serve a valuable role in high-intensity training for top candidates. It is a desperately needed parallel system. It is the system we should have built yesterday if we truly believe our own timelines.

This Is Your Sign

Let’s be blunt. If your P(doom) is high and your timelines are short, waiting is an indefensible strategy. Waiting for the right fellowship, the right grant, the right moment—it’s a gamble against a clock that might be ticking faster than we think.

The belief in short timelines is a call to action. It is a scream that demands urgent, desperate, and unconventional measures. It demands that we stop thinking in terms of who we can filter out and start thinking in terms of how we can onboard everyone who is capable and willing to help.

If you are one of the thousands who has been sitting on the sidelines—a programmer, a manager, a writer, a researcher—feeling the urgency but seeing no clear path forward, this is your sign. The cavalry isn’t coming. We are the cavalry.

We need all the heroes we can get.