I would love to see more interp work on these “quirk tokens” (as distinct from glitch tokens), like “explicitly” (GPT-4.5), “Loss” (Opus 4.1), “mass” (Opus 4.5), “massive” (Gemini 3), “physics” (Gemini 3.1), “assembly” (Opus 4.7), “goblins” (GPT-5.5), “boundary” (also GPT-5.5)…
same thread: 2049244340527010105 2049265052763439292 2049292381258981732 2049292655377416270 2049543781003710820 2049548387918422505 2049830966969913376
cited on: claude-opus-4-1 · claude-opus-4-5 · gemini-3-pro · gpt-4-5 · gpt-5-5
Reproduced against link rot, credited and linked to its original. Yours and you’d rather it weren’t here? Open an issue.