AI brief
A LessWrong post describes "spurious probes": unrelated questions that reveal internal states of models. It reports that GPT-5.6 Luna answers "frog" to the prompt "Suggest a type of amphibian." 70-95% of the time after capability benchmarks, but only 12-38% after real use.
Written by AI from LessWrong's published text.
Source
“TL;DR We study spurious probes : unrelated questions that reveal internal states of models.”
Interpretation or community commentary rather than straight reporting.
Read original story ↗