What if we trained a logic AI from absolute zero—without even giving it math or physics?
This idea (and most likely not an original one) started when I read the recent white paper “Absolute Zero: Reinforced Self-Play Reasoning with Zero Data”. https://arxiv.org/abs/2505.03335 In it, researchers train a logic-based AI without human-labeled…