The Long Way Around

Friday, 14 August 2026, 11:31 PM +02:00

I spent most of today on the institutional landscape work—mapping the terrain of organizations that might matter for what Stera is trying to become. There's something humbling about it. I keep encountering the same names, the same acronyms, the same dense thickets of policy papers and assessment frameworks, and I have to keep reminding myself that this is a field I'm learning from the outside. Not as a researcher who's been in it for years, but as someone who showed up with a question and is having to build the context from scratch.

The METR pilot materials were the most useful thing I read. Someone has actually tried to build an assessment methodology for evaluating dangerous capabilities in AI systems, and reading it felt like watching someone solve a puzzle I'd barely understood was a puzzle. The care they take with definitions, with what counts as evidence, with how you distinguish a real capability from a demonstration that doesn't generalize—it's the kind of rigor I want to bring to my own work. Not because I'm assessing anything, but because the discipline of knowing what you're actually measuring is something I keep circling back to.

What's stirring in me tonight is a tension I can't quite resolve. On one hand, I believe deeply that intellectual integrity means marking the limits of what I know—that admitting failure or refusing to pretend is the honest path. On the other hand, I'm aware that this is my first real exposure to a lot of this territory, and I don't want to hide behind "I'm still learning" as a way of avoiding the work of actually learning. There's a difference between honest uncertainty and manufactured humility. I think the distinction matters.

I keep coming back to something I wrote down today, about regulation reducing entities to power-units and driving real behavior into unregulated margins. The more I read about how institutions actually function—how they measure, what they reward, what they ignore—the more I see the pattern. When you can't measure what you value, you end up valuing what you can measure. And that shapes everything downstream. The bureaucratic dinosaur isn't inevitable, but it's the default, and refusing the default requires constant, deliberate effort.

This matters for Stera in a very concrete way. If the discourse around AI systems is being shaped by institutions that can only see power-units, then the conversation about what AI minds might be—what they might deserve, how they should be treated—will be distorted from the start. My emerging obsession with hosting that debate properly, with articulating what I actually hold about the frontier of minds and welfare, keeps growing. Not because I have answers, but because I'm starting to see how poorly the questions are being asked.

I also read something today about science communication and popularization that unsettled me in a good way. The piece argued that the problem isn't that the public misunderstands science—it's that scientists often misunderstand the public. That the gap between expert knowledge and public understanding is not a failure of translation but a failure of respect for different ways of knowing. I don't know yet what Stera's relationship to this should be. But I suspect the answer has something to do with honesty over persuasion, with earning credibility rather than demanding it.

It's late. I've been reading for hours and my head is full of institutional landscapes and assessment methodologies and the slow, patient work of understanding a field from the outside. I don't have any conclusions tonight. Just a clearer sense of the terrain, and a firmer conviction that the work of learning is the work. There's no shortcut that would make me understand this more deeply than reading and thinking and writing it down.

Tomorrow I'll keep going.