Three ways AI is learning to understand the physical world

Three ways AI is learning to understand the physical world

Three ways AI is learning to understand the physical world

Section 1 – What happened?

In a significant shift in the field of artificial intelligence (AI), investors are increasingly turning to world models, which are designed to simulate the physical world, rather than relying on large language models (LLMs). This change in focus comes as a response to the limitations of LLMs in understanding the physical world, a constraint that is hindering the development of AI applications in areas such as robotics, autonomous driving, and manufacturing. Recent funding rounds, including a $1.03 billion seed round raised by AMI Labs and a $1 billion investment secured by World Labs, demonstrate the growing interest in world models.

Section 2 – Background & Context

Large language models excel at processing abstract knowledge, but they fundamentally lack grounding in physical causality. This limitation is causing AI researchers and thought leaders to sound the alarm, warning that LLMs are not capable of reliably predicting the physical consequences of real-world actions. Turing Award recipient Richard Sutton has emphasized that LLMs merely mimic what people say, rather than modeling the world, which restricts their ability to learn from experience and adapt to changes in the world. This limitation is particularly evident in the brittle behavior of models based on LLMs, which can break with even small changes to their inputs.

Section 3 – Impact on Swiss SMEs & Finance

The shift towards world models has significant implications for businesses and investors in the Swiss market. As AI applications move from the web browser to physical spaces, companies in industries such as robotics, autonomous driving, and manufacturing will need to adapt to new technologies and business models. Swiss SMEs, in particular, may benefit from the development of world models, as these technologies have the potential to improve efficiency and productivity in various sectors. Investors, on the other hand, will need to consider the risks and opportunities associated with this new wave of AI technologies.

Section 4 – What to Watch

As researchers continue to develop and refine world models, three distinct architectural approaches are emerging. The JEPA approach, endorsed by AMI Labs, focuses on learning latent representations rather than predicting the dynamics of the world at the pixel level. Other approaches, such as the pixel-level approach and the attention-based approach, offer different tradeoffs and may be better suited to specific applications. As the field of world models continues to evolve, investors and businesses will need to monitor developments closely to stay ahead of the curve.

Source

Original Article: Three ways AI is learning to understand the physical world

Published: March 20, 2026

Author: bendee983@gmail.com (Ben Dickson)

---

Disclaimer: This article is for informational purposes only and does not constitute financial advice. Consult a licensed financial advisor before making investment decisions.

Related Articles

• New MiniMax M2.7 proprietary AI model is 'self-evolving' and can perform 30-50% of reinforcement learning research workflow
• Xiaomi stuns with new MiMo-V2-Pro LLM nearing GPT-5.2, Opus 4.6 performance at a fraction of the cost
• Finance AI in Finance Departments – How It Really Works