Murad Farzulla
Writing about what the work
actually did.
Papers report conclusions. This is the other half: the check that killed a result, the control that changed what a number meant, the retraction I had to issue against my own formalism. Finished research lives at dissensus.ai. This is the reasoning.
Essays
-
AI Friendship ·
Conversations Are Not People
I have a paper arguing that friendship is substrate-independent. Going looking for the prevalence data that would tell us whether it matters at scale, I found two good public measurements, nearly differenced them (they use incompatible taxonomies), and then hit the larger problem: they count conversations, and the question is about people.
-
Retraction ·
The Arm That Was Never Adversarial
I published a closed-form expression for friction in delegated systems. Trying to derive it rather than fit it took three independent passes and left two of six claims standing. The most useful failure was an experimental arm that was symmetric under the very thing it was built to vary.