To effectively use AI, one must understand the physics behind the problems being solved, allowing us to design AI that can provide answers directly applicable to real-world scenarios.

The design space for physical products is vast, requiring careful sampling strategies to ensure that AI models learn effectively from both dominant designs and edge cases.

The main challenge in developing AI for physical engineering is the lack of sufficient training data, as there haven't been millions of airplane designs since the Wright brothers.

Currently, we manually build the component catalog for synthetic data generation, but we are developing AI tools to automate this process and intelligently assemble components into systems.

AI applications that ingest real-world sensor data and robotics are crucial building blocks for achieving physical AI, which is essential for developing engineering AGI.

The challenge in AI lies in balancing supply and demand; while we invest in building models, we must also focus on how to effectively utilize these models for economic and mission advantages.

Achieving sustainable scale in AI applications requires aligning models with real-world use cases and optimizing user behavior throughout the product design process.

To create synthetic data sets for complex systems, we need a comprehensive catalog of components that is significantly larger than the typical system design, potentially requiring hundreds of millions of parts.

Synthetic data plays a crucial role in drug discovery, especially in chemistry, where we can use theoretical models to create scalable molecular simulations and enhance the information content of our datasets.