Turing Post (Ksenia Se)· @TheTuringPost · X·· 1 天前AI 评分22
AI 导读
修复 surprisingly small。“Natura non facit saltus”——自然无跳跃。 当行动与扰动作为独立力作用时,其相互作用在时间上是二阶的。因此: 假设结果 ≈ 观测结果 + 预测(其他行动) − 预测(已采取行动) 两次调用同一预测器并做一次加法。无需反事实数据、额外参数或重新训练。他们证明了 O(h²) 误差界,且仅使用非配对事实数据的方法无法在速率上做得更好。
正文 · 原文
The fix is surprisingly small. "Natura non facit saltus" – nature makes no leaps.
When the action and the disturbance act as separate forces, their interaction is second-order in time. So:
what-if outcome ≈ observed outcome + prediction(other action) − prediction(action taken)
Two calls to the same predictor and one addition. No counterfactual data, no extra parameters, no retraining. They prove an O(h²) error bound and that no method using only unpaired factual data can do better in rate.
来源:Turing Post (Ksenia Se) · x.com