Unverified50% confidenceTradeoffExact time

剩余12%的失败率主要来自复杂参数组合、边界情况或格式偏差，对生产环境是不可忽视的挑战

1

Sources

50%

Confidence

Long-term

Relevance

7/15/2026

First Seen

Sources

GPT-2 Fine-Tuning Experiment: 88% Function Calling Success Rate with a 355M Small Model

Related Claims

Unverified传统ReAct模式在复杂场景下存在瓶颈，长链路推理中的误差累积效应会导致任务失败率上升70% similar Unverified引导式生成模型面临数据依赖挑战，极端事件数据本就稀缺，数据质量直接决定估算可靠性70% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/513757

API

curl https://kongchang.com/api/v1/knowledge/claims/513757

MCP

get_claim(id=513757)