Skip to main content
Back to catalog

ICENSE

The word “LICENSE” minus its first letter — an artifact of open-source license text flooding the corpus.

Raw token

"ICENSE"

Observed behaviors

  • 01In GlitchMiner's demo, Mistral-7B-Instruct-v0.3 silently “corrected” it to LICENSE.arXiv
Observed or documented on
Mistral
Behavior types
Unspeakable
Tokenizer
Mistral BPE
Discovered by
wooozihui 等 (GlitchMiner)

References