Two pieces of writing this week, separated by genre and readership but connected by diagnosis. An arXiv paper by Gordon Burtch titled 'The Hitchhiker's Guide to Monoculture' argues that large language models produce homogeneous outputs that flatten the diversity of ideas when AI coding assistants and content tools are widely adopted. An Artnet op-ed by Adam Levine argues that museums' institutional insularity, the tendency to hire only from within the curatorial track, is similarly suppressing the field's intellectual vitality. Different systems. Same failure mode: closed loops that reward sameness.
Homogenization Is an Architectural Choice
Burtch's paper is specifically concerned with what happens when AI tools built on the same underlying models are used by millions of developers simultaneously. The diversity of solutions collapses. Innovation becomes variation within a narrow band. This is not a bug introduced by bad actors; it is an emergent property of shared optimization targets. The art market parallel is almost too clean: Christie's just reported a $4.5 billion first-half haul, its biggest in five years, driven by the collections of S.I. Newhouse and Agnes Gund. Blue-chip provenance, institutional names, legacy collections. The market optimizes for what it already recognizes, just as the LLM optimizes for what its training data already contains.
The Talent Pipeline as Training Data
Levine's argument in Artnet is that museum leadership that draws exclusively from the curatorial track produces institutions with a narrow range of problem-solving strategies. He wants cross-pollination from the art market side, people who understand liquidity, negotiation, and the mechanics of private taste formation. Kyle Chayka's analysis of algorithmic homogenization, explored in his Culture Slop conversation, provides the unifying frame: when any system, algorithmic or institutional, is optimized for engagement or consensus, it systematically deprioritizes the outlier. The outlier is where the new thing comes from. Both museums and LLMs are, right now, building elegant machines for producing more of the same.