arxivcs.AI2026-07-31
AMTFV: Agentic Mathematical Tool-Flow Verification for LLM Self-Correction
Rui Zou, Yutao Zhu, Mengqi Wei, Ji-Rong Wen
Large language models have demonstrated strong mathematical problem-solving capabilities, yet reliably verifying their candidate answers remains challenging. Existing representative methods mainly revise outputs through natural-language reflection or assist verification by direct…