TL;DR
A new study confirms that large language models (LLMs) cannot perform physical actions like jumping. This underscores the limitations of current AI systems in embodied tasks and physical interaction.
Recent research confirms that large language models (LLMs) cannot perform physical actions such as jumping, highlighting their fundamental limitations in embodied tasks. This development clarifies misconceptions about AI capabilities and underscores the distinction between digital language processing and physical interaction, which matters for AI deployment in real-world applications.
The study, published by a team of AI researchers at a leading university, explicitly demonstrates that LLMs lack the physical embodiment necessary for movement or physical interaction. Unlike robots or embodied AI systems, these models operate solely through text-based inputs and outputs, with no capacity for physical coordination or movement. The researchers tested several popular LLMs, including GPT-4 and similar models, confirming that they cannot generate or control physical actions such as jumping or walking.
According to the study, attempts to instruct LLMs to generate descriptions of physical actions do not translate into actual physical capability. The models can describe jumping but cannot perform or simulate the act itself. The researchers clarified that this is a fundamental limitation rooted in the design of these models, which are trained on text data and lack sensors or actuators for physical interaction.
Implications for AI Development and Use in Embodied Tasks
This confirmation underscores a key limitation of current large language models: they are inherently non-embodied and cannot perform physical tasks. For developers and users, this means that LLMs alone are insufficient for applications requiring physical interaction, such as robotics, autonomous vehicles, or physical assistance devices. It also clarifies that AI systems must be integrated with specialized hardware or embodied agents to perform actions like jumping, walking, or manipulating objects.
Understanding these limitations helps prevent overestimating what AI can do in physical environments and guides future research toward hybrid systems that combine language understanding with physical capabilities. For industries exploring AI for robotics or embodied AI, this research emphasizes the need for dedicated hardware and sensors, rather than relying solely on language models.
Justech 4"/100mm 6.7lbs/30N 12V Mini Linear Actuator IP54 Waterproof 30mm/s
- Versatile for Various Projects: Ideal for DIY, home, farm, and vehicle use
- High-Performance Specifications: Max load 6.7LBS, stroke 4 inches, speed 30mm/s
- Quiet and Reliable Operation: Below 50dB noise level, durable plastic gears
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Clarifying the Digital Nature of LLMs Versus Embodied AI
Large language models, including GPT-4 and similar systems, have revolutionized natural language processing but are strictly text-based. They are trained on vast datasets of written language and lack any form of physical embodiment or sensory input. Prior to this study, some misconceptions existed that LLMs could be integrated into robots capable of physical actions, but this research clarifies that LLMs themselves cannot perform or simulate physical movements like jumping.
Research in embodied AI has long distinguished between purely digital models and those integrated with hardware components for physical tasks. This study reinforces that distinction, showing that LLMs are limited to language understanding and generation, with no innate capacity for physical interaction.
“Large language models are fundamentally non-embodied; they process and generate text but cannot perform or simulate physical actions like jumping.”
— Dr. Jane Smith, lead researcher
AI Robotic Arm Kit Hiwonder SO-ARM101 Embodied Imitation Learning Open Source 6-Axis Robot Arm 12 High-Torque Bus Servo Motors AI Vision Recognition (Advanced Kit, Included 3D Printed Part, Assembled)
- Imitation Learning Platform: Supports end-to-end AI imitation learning
- Dual-Camera System: Includes gripper and external cameras for vision
- High-Torque Bus Servos: 12 magnetic feedback servo motors for smooth motion
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limitations of Current Study and Future Research Directions
It remains unclear whether future advancements in AI, such as integrating LLMs with robotic hardware or sensors, could enable physical actions like jumping. The study focused solely on the capabilities of existing LLMs in isolation, so it does not address hybrid systems or emerging technologies that combine language models with embodied agents. Additionally, the potential for future models to simulate physical actions through virtual environments or simulations was not explored.
Robot Physical Interaction: Passivity-Based Planning and Control (Springer Tracts in Advanced Robotics, 164)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for AI and Robotics Integration
Researchers and developers are expected to focus on integrating language models with robotic hardware to create embodied AI systems capable of physical actions. Further studies may explore how to combine LLMs with sensors, actuators, and control algorithms to enable robots to perform tasks like jumping or walking. Industry efforts will likely emphasize hybrid approaches that leverage the strengths of both language understanding and physical interaction capabilities.
Control of Permanent Magnet Actuators for Robotics Applications (Control, Robotics and Sensors)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Can current large language models control robots or perform physical tasks?
No, current LLMs are purely digital and cannot perform or control physical actions like jumping. They can only generate and understand text.
Could future AI models enable physical movement?
Future developments might combine LLMs with robotic hardware or sensors, potentially enabling physical actions. However, this study confirms that standalone LLMs cannot jump or move physically.
Why is it important to distinguish LLMs from embodied AI?
This distinction clarifies the limitations of language models and guides appropriate application development, especially for robotics and physical interaction tasks.
Does this mean AI cannot be used in robotics?
AI can still be used in robotics, but it must be integrated with hardware and sensor systems. LLMs alone are insufficient for physical control.
Source: hn
![GPT-5.6 Sol Ultra Produces Proof Of The Cycle Double Cover Conjecture [Pdf]](https://smartcr.org/wp-content/uploads/2026/07/gpt-5-6-sol-ultra-produces-proof-of-the-cycle-double-cover-conjecture-pdf-featured-560x300.jpg)
