Unverified50% confidenceFactExact time
GPT-4V 等多模态模型在直接输出屏幕坐标时存在显著的空间定位偏差
1
Sources
50%
Confidence
Long-term
Relevance
7/18/2026
First Seen
Sources
Related Claims
UnverifiedIn a 3D scene generation test, GPT-5.2 Codex produced spatial positioning errors such as a television embedded in the floor, while Opus 4.5 delivered a stronger overall atmosphere and higher 3D scene fidelity.75% similarUnverifiedExperiments show that the performance of multiple mainstream models, including GPT-4, exhibits a clear U-shaped attention curve.75% similarUnverified早期 GUI Agent 方案依赖计算机视觉截图分析（如 GPT-4V），存在延迟高、定位不精准的问题73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/553649API
curl https://kongchang.com/api/v1/knowledge/claims/553649MCP
get_claim(id=553649)