AI 学空间，只练一种题会锁死自己
训练一个空间 AI 去回答「物体在左边还是右边」，它并不需要真的记住距离；反过来，只练距离估计，左右关系又可能完全没约束。研究者发现，当后训练只盯一种输出，模型会「锁死」在那种问法上，换一种问法就变差——他们叫它 interface lock-in。解法是把不同种类的空间监督按时间顺序重放，像给 AI 补历史课，同一个模型在三个空间基准上同时变好，其中一个涨了 13.41 分。它不是你明天能用上的东西，但解释了为什么 AI 有时换个问法就答错：不是它不懂，是它只练了那一种表达。
📄 原文摘要(英文)
Training a spatial VLM to answer which side an object lies on does not require it to preserve the metric displacement that determines that answer, while supervision for a distance estimate may leave categorical relations largely unconstrained. When post-training emphasizes one readout, improvements can therefore concentrate on that interface while transfer to others remains weak or even degrades; we call this observed specialization interface lock-in. ChronoReplay addresses it by replaying heterogeneous spatial supervision as chronologically ordered episodes, retaining source-native targets, and balancing supervision across families. Under the same observation access, a single ChronoReplay checkpoint improves OVO-S-Bench, OST-Bench, and UCS-Bench together, including a 13.41-point gain on OST-Bench, and achieves the highest three-benchmark mean among the supervision settings compared here. Same-data controls further separate the ingredients of this transfer: answer serialization redistributes gains across OST and UCS, family-balanced learning improves over ordinary joint SFT and natural turn weighting, and chronological replay exceeds a shifted-order control by about 2.41 OST points across two paired seeds. Adaptive frame selection then changes evidence access while keeping the trained model fixed, providing an additional inference-time gain at lower visual cost. Streaming spatial performance depends on what distinctions supervision preserves, how heterogeneous histories are organized during learning, and what causal evidence inference makes available at test time.