There’s a particular kind of vertigo that comes with running a non-profit. Every morning, I wake up knowing exactly what we’re trying to achieve – helping advanced AI go well for humanity by training the next generation of policymakers – but having only rough hints as to whether or not we’re succeeding.

This isn’t a problem I faced running a start-up. We had the luxury of clarity: revenue growth, EPC, dashboards of customer acquisition costs. Every day was full of metrics and data points that clearly signalled whether we were on the right path. And if we ever dropped the ball on service quality our customers were all too happy to let us know.

At Talos, measuring our success is more art than science. We can count the number of fellows we’ve trained and follow their careers closely, but we’ll never really know the scale of our counterfactual impact.

The main reason for this is the lack of market signals telling us we’re doing a good job. There’s just no direct connection between delivering on our mission and keeping the lights on. This sets up an incentive structure where we might optimise for what’s measurable and presentable rather than what’s most effective.

This isn’t just philosophical hand-wringing. The metrics we choose shape our strategies, our resource allocation, and ultimately, our theory of change itself. When we focus too narrowly on measurable outcomes, we risk missing the forest for the trees – or worse, cultivating a carefully manicured garden of metrics while the broader ecosystem of impact withers.

Right now this is just an open ambiguity to keep an eye on and to manage as best we can. I’m sure this is a fact of life for many organisations in the non-profit sector, and all we can do at Talos is constantly try to live up to our lofty ideals.