TY  - RPRT
TI  - Uncalibrated Reasoning: GRPO Induces Overconfidence for Stochastic Outcomes
AU  - Michael Bereket
AU  - Jure Leskovec
PY  - 2025
UR  - https://arxiv.org/abs/2508.11800
ID  - 2508.11800
ER  -