hraness

saved

The Hofmann Wobble

by Ben LernerHarper's Magazinepublished

gist

Ben Lerner’s autofictional Harper’s essay reconstructs a mid-2000s Wikipedia campaign. A “New Media Fellow” built sockpuppet editors, farmed authority through fly-fishing busywork, and reframed pages on tax policy, war, and climate. The later self-critique is that Wikipedia became default memory while remaining writable; putting facts on a screen off-loaded them from people; treating the site as both encyclopedia and news source canceled the slowness memory needs. The piece ends by handing the last word to ChatGPT.

ideas

  • Sockpuppets beat a named activist. A single institute account got reverted. MormonCricket and allied voices with long service histories could reframe tax, war, and climate pages and defend them on talk pages.
  • A page can mint a fact. A ten-minute Shinwar Massacre entry entered NPR speech. Bocce and Poe pranks later recirculated as sourced claims.
  • The screen off-loads memory. Treating Wikipedia as both encyclopedia and news source cancels the slowness remembering needs. Displaying a fact does not put it in a reader.
  • Facts do not settle tribes. Lerner later doubts that documenting greenwashing or secret prisons changes belief, and that a denial frame can disguise willful extraction as ignorance.
  • The encyclopedia is becoming training data. He stops making pages because language models already generate the bland prose the New Media Fellow spent years pasting in.

quotes

I could make edits through ten or eleven fictional users and swarm talk pages to defend them when necessary.

Ben Lerner, describing the sockpuppet editorial network.

Such edits were somewhere between childish pranks and tiny terrorist attacks on the historical record.

Ben Lerner, on planting stray facts to see whether they linger.

The NMF was prominently displaying what needed to be remembered in a medium antithetical to (human) memory.

Ben Lerner, stating the essay’s judgment of Wikipedia as memory.

Untrue crickets give way to bots.

Ben Lerner, closing the Wikipedia era as training data for language models.