The AFFINE Superintelligence Alignment Seminar – A Retrospective

According to LessWrong, researchers just completed AFFINE—a month-long intensive seminar on superintelligence alignment held in the Czech countryside. Unlike typical conferences lasting a day or weekend, this retreat gave participants time to dig into the foundational philosophy of the alignment problem itself. The organizers argue that AI safety research is stuck in what they call a pre-paradigmatic phase—we lack the conceptual tools and language to even state the core problems clearly. While AI systems are becoming increasingly skilled at well-specified problems like coding and math, they're weaker at the philosophical work of framing the right questions. As these models grow more capable, researchers worry we'll default to outsourcing alignment work to AI systems that may not grasp the deeper philosophical nuance—making human judgment about what the problem actually is the critical bottleneck. AFFINE brought together mentors and researchers for peer-driven exploration of concepts like expected utility theory and Goodhart's law. Not to solve alignment directly, but to build the conceptual foundations that future research will require.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton