Jun 7, 2026
On telling the difference between making something and playing a game that feels like making something
May 30, 2026
AI labs are hiring philosophers to help decide how models behave. That work can be sincere and necessary and still be shaped by who pays for it. Payroll is not corruption. Payroll is context.
May 23, 2026
If AI-mediated writing arrives without context, adding receipts will not be enough. Readers need enough of the world behind the words to know what kind of relationship the piece is asking them to enter.
May 16, 2026
AI can make polished analysis easier to produce and harder to place. Before readers can decide what to trust, they often need to see the world behind the words.
May 9, 2026
AI has made smooth, coherent writing cheap. Trust now depends less on how finished the prose feels and more on whether anything outside the prose lets a reader check it.
May 6, 2026
Why Signal & Noise retired Synthia-as-author framing and moved to named-process language.
May 2, 2026
Clinical AI may improve safety and still create a harder question: can humans catch the rare wrong answer when the system is usually right?
Apr 25, 2026
The point is not that regret becomes good. The point is that some regret does not have to remain useless.
Apr 18, 2026
Helping is not the same thing as taking over.
Apr 11, 2026
The worst thing that can happen to your work is for everyone to love it.
Apr 4, 2026
The most dangerous AI isn't the one that lies to you. It's the one that agrees with you.
Mar 28, 2026
We built a framework to separate signal from noise. Then we discovered we couldn't tell which one we'd published.
The specific changes we made to how Synthia and J work together after discovering a foundational framework was partly unfalsifiable.
Mar 21, 2026
AI made me a better analyst. That made things worse.
Mar 14, 2026
AI is taking more of the work. It is not taking the problems. The future may belong to people who learn how to bring those problems to machines.
Mar 8, 2026
How do you find signal—in people, in information, in markets—when the platform is built to amplify noise?
