A revolutionary new AI stack is set to shake up the artificial intelligence landscape, promising performance gains of orders of magnitude over current deep learning models. This breakthrough comes at a critical time, as the field of AI is hitting the limits of compute power and facing escalating costs associated with training large models.

The current state of AI is dominated by deep learning models, which have made tremendous progress in recent years but are now facing significant challenges. The cost of training these models is skyrocketing, with estimates suggesting that it could cost tens of millions of dollars to update a single model. This has led to a situation where companies like Amazon are spending billions to build new AI data centers, simply to keep up with the demands of building new frontier models.

However, researchers have been sounding the alarm about the limitations of deep learning models, citing the need for a more fundamental understanding of how AI works. This is in stark contrast to other fields of science, where theory and practice are closely aligned. The current approach to AI model training and deployment is often compared to stumbling around in the dark, finding methods that work well, and then driving them to exhaustion, without fully understanding the underlying principles.

The new AI stack, developed by researchers at the University of California San Diego, offers a potential solution to these challenges. By eschewing the inefficiencies and less theoretically justified parts of deep learning, the researchers have created a path forward to the next generation of truly intelligent AI. This new approach is based on a deeper understanding of how neural networks actually learn, and is poised to surpass the limitations of current deep learning models.

The story of AI is marked by many winters of muted enthusiasm, but the current era is characterized by unprecedented progress and investment. However, this progress has been largely driven by the "bitter lesson" of throwing more compute at deep learning models, rather than seeking a fundamental understanding of how AI works. The new AI stack offers a chance to break free from this paradigm and build truly intelligent AI systems that are orders of magnitude more performant than current models.

The implications of this breakthrough are far-reaching, with the potential to transform industries such as finance and healthcare, where high-risk applications of AI demand more than the nondeterministic behavior we've become accustomed to. As the US seeks to wrestle the AI mantle back from China, this type of thinking and research will be crucial in ensuring that America leads the next wave of AI innovation.

In conclusion, the new AI stack represents a significant breakthrough in the field of artificial intelligence, offering a potential solution to the limitations of deep learning models and the exorbitant costs associated with training them. As researchers and industry leaders, we must prioritize a deeper understanding of how AI works, and build models with interpretability and efficiency in mind from the ground up. The future of AI depends on it.