Simple penalty method improves AI reasoning efficiency

Original: probably the best reward function for reasoning efficiency i've seen

Source: x.com ↗