Qualcomm CEO says Nvidia’s Huang is "100% right" about future of AI #AI #Nvidia

Physical AI: The Next Wave

Key Concepts: Physical AI, Large Language Models (LLMs), Sensor Data, Robotics, Humanoid Robots, Training Data, Embodied Intelligence.

This discussion centers on the assertion that the next significant advancement in Artificial Intelligence will be “Physical AI,” a field poised to be even more impactful than current focuses on robotics, particularly humanoid robots. The speaker agrees with a previous point made by “JSN” regarding this impending shift.

The Limitations of Current AI – Large Language Models

The core argument presented is that current AI development is heavily concentrated on Large Language Models (LLMs). These models, like those powering many chatbots and text generation tools, are trained on vast datasets of text and internet information. This training allows them to mimic human language and thought processes. However, the speaker emphasizes this is fundamentally limited. LLMs operate within the realm of information representation – they process and generate text based on patterns learned from existing data.

The Potential of Physical AI – Learning from Sensation

Physical AI, in contrast, is defined by its training on sensor data. This means the AI learns not from text, but from direct experience – what it “sees,” “senses,” and interacts with in the physical world. This is a crucial distinction. The speaker doesn’t elaborate on specific sensor types at this point, but the implication is a broad range including visual, auditory, tactile, and potentially others.

Beyond Humanoid Robots: A Broader Scope

While humanoid robots are often the first image that comes to mind when discussing physical AI, the speaker argues that the field’s potential extends far beyond them. Humanoid robots represent an application of physical AI, but not the entirety of its scope. The speaker suggests the opportunity is “even bigger than that,” implying a wider range of applications leveraging AI’s ability to learn from and interact with the physical environment.

Embodied Intelligence & The Importance of Data Source

The underlying principle driving this perspective is the concept of embodied intelligence. This suggests that intelligence isn’t simply about processing information, but about being situated within a physical body and interacting with the world. The type of data used for training is therefore paramount. LLMs are trained on representational data (text), while Physical AI is trained on experiential data (sensor data). This difference in data source fundamentally alters the nature of the intelligence developed.

Logical Connection & Future Implications

The speaker establishes a clear contrast between the current trajectory of AI development (LLMs) and a predicted future trajectory (Physical AI). The argument isn’t that LLMs are unimportant, but that Physical AI represents a qualitatively different and potentially more transformative advancement. The focus on sensor data suggests a move towards AI systems that can operate more autonomously and effectively in real-world environments, going beyond simply understanding and generating language.