Commonsense Reasoning in Computer Vision
The survey provides a comprehensive overview of methods to enhance AI's understanding of everyday scenarios through commonsense integration.
The survey provides a comprehensive overview of methods to enhance AI's understanding of everyday scenarios through commonsense integration.
Gary Marcus 说 ARC-AGI-3 上跑得最好的都用了神经符号世界模型,想看这方向为什么牛,可以读这条。
Gorard团队实测了GPT-5.6、Fable 5等模型,发现它们连基础非线性PDE都解不对,而Lanyon用更少token就能搞定,想了解模型短板可以看看。
这项研究解决了高风险AI应用中模型过度自信的痛点,做可解释AI或安全关键系统的团队可以直接参考COCOCO框架来提升预测可靠性。