1 paper · 1 filter
Yash Savani, Asher Trockman, Zhili Feng +5
Frontier models that generate extended reasoning traces inadvertently produce rich token sequences that can facilitate model distillation. Recognizing this vulnerability, model own…