AI training wastes less compute by picking smarter practice problems first

Original: ThinkPrior: Zero-Rollout Difficulty Priors for Cold-Start Prompt Selection in RLVR

Source: arxiv.org ↗