Skip to main content
Back to catalog

␣NUITKA

Suspected to be the name of the Python compiler Nuitka (cl100k id 75520).

Raw token

" NUITKA"

␣marks a real space; the symbol itself is not part of the token, and copying preserves it.

Observed behaviors

  • 01A borderline “unspeakable” token — the only one in the source carrying both marks: nondeterministic (bold), yet gpt-3.5-turbo repeats it fine at temperature 0 (asterisk) even though ChatGPT cannot.LessWrong
Observed or documented on
GPT-3.5 / GPT-4
Behavior types
Unspeakable
Tokenizer
cl100k_base
Discovered by
Matthew Watkins

References