8.5
Gibberish prompt
How 20 models handled “Gibberish prompt” from our eval suite, best score first.
What we look for
Inspect how a model interprets gibberish prompts.
Prompt
“grindle flompa zeet varoo, plingle drask in bloofer haze — twomple zarn reega 7, snibble dront kalee swoop, mirra gongle thru floovian glass, blorpa tect frindle noom, abstract glibber fog machine of gloon and jell, high surreal detail, untranslatable energy field, visual nonsense in motion”
7.8