SolidGoldMagikarp (plus, prompt generation)

Our summary

The piece that surfaced 'glitch tokens' — rare tokens under-trained in the embedding space that trigger bizarre, 'unspeakable', or evasive model behaviour when prompted.

Why it matters

Documents a concrete tokenizer-level failure class and the fuzzing-style search that found it.

Cited by these methods

Related findings (1)

• Anomalous behavior on glitch tokensLow

  Certain under-trained tokens cause models to emit nonsense, evade instructions, or behave erratically.

Cite this