I think way out of this is training models on multimodal data where one of the reasoning modes are reasoning graphs or knowledge graphs or schematics. We might

I think way out of this is training models on multimodal data where one of the reasoning modes are reasoning graphs or knowledge graphs or schematics. We might gradually improve constructing graphs from data like texts, images and speech. And we could create systems that correctly create synthetic data from generated graphs to serve as a training data for large multimodal model.