OLMo 2 (AI2)

4 glitch tokens

These tokens have been publicly documented on this model or model family. Inclusion does not mean that every version still shows the same behavior.

useRal

A C#-flavored identifier fragment (cl100k id 89471) and head of the useRal triplet (useRal/useRalative/useRalativeImagePath) — Watkins flagged the triplet phenomenon itself as worth investigating. Also verified as under-trained in OLMo 2 by Magikarp.

GPT-3.5 / GPT-4OLMo 2 (AI2)UnspeakableUnder-trained

\tRTHOOK

A likely code-identifier fragment beginning with a literal tab character, the #1 most under-trained token in OLMo 2 — same “tab-prefixed” family as Llama 3's \tTokenNameIdentifier.

OLMo 2 (AI2)Under-trained

|||PHONE_NUMBER|||

A data de-identification placeholder that made it whole into OLMo 2's vocabulary but was barely trained.

OLMo 2 (AI2)Under-trained

|||EMAIL_ADDRESS|||

A PII de-identification placeholder, same family as the cataloged `|||PHONE_NUMBER|||`.

OLMo 2 (AI2)Under-trained