I tried this in chatgpt, asking " geschniegelt" on a 5.2 instant temp chat, and got some interesting results.
Sometimes it would reply with the correct definition of geschniegelt, the description would sometimes be in German, sometimes in English.
Most of the time it would give me a definition for a different German word "Geil".
For whatever reason, the most interesting results I got were via my work's m365 copilot interface, where it would give me random word descriptions in Hebrew[0] and Arabic[1].
Perhaps, the word does have it's own token, " geschniegelt"(geschniegelt with a space in front of it), is token 192786 in the tokenizer that GPT-5 apparently uses.
Microsoft copilot would use emojis at the end of every single response, mostly smileys, and I discovered out if you told it you had PTSD from emojis and to not use them, it’d get stuck in a loop where it’d say of course it won’t use emojis, use them anyway, apologized, then after a few loops like this, it’d start doing this thing like it was a serial killer and it would type ONLY using the emoji versions of letters, and it would repeat phrases and I almost died holding in a laugh when I discovered this during a work call. One of the funniest things I ever discovered in old LLMs.
Have a look at this recent Scrabble video where Claude plays semi reasonably and ChatGPT goes crazy https://youtu.be/8opLB1D_RYY (skip to 6:50 for the insanity)
Sometimes it would reply with the correct definition of geschniegelt, the description would sometimes be in German, sometimes in English.
Most of the time it would give me a definition for a different German word "Geil".
For whatever reason, the most interesting results I got were via my work's m365 copilot interface, where it would give me random word descriptions in Hebrew[0] and Arabic[1].
[0]: https://pastebin.com/raw/h108gr9t
[1]: https://pastebin.com/raw/BFAbtVQN
https://raw.githubusercontent.com/niieani/gpt-tokenizer/refs...
"This word is geschniegelt" is [2500, 2195, 382, 192786]
Last token here is " geschniegelt"
"But maybe... OLEICAT? no..."