Ankit Maloo argues that AI models are improving at mathematics by pursuing uncertain approaches through lengthy calculations, with verification and parallel agents supporting the process. He proposes that problems requiring fewer conceptual leaps and narrower search trees are more tractable for models, offering a way to think about which problems they may be able to solve.
Maloo says human experts can help by narrowing the search trees models must explore. He also suggests that access to failed attempts could support mathematical understanding beyond what the final proof conveys.
AI 根据报道生成 · 24 分钟前更新
