Co-founder of OpenAI: Autopilot and VR are both "misguided" AI agents are the future

Author: Ge Jiaming

Ge Jiaming

After Huang Renxun and Musk saw the development potential of AI agents one after another, OpenAI co-founder and former Tesla director of artificial intelligence Andrej Karpathy also recently shouted that **AI agents represent a crazy future. **

Andrej Karpathy bluntly said that he was "distracted by autopilot" when he was working at Tesla, and studying autopilot and VR is not the right way to develop AI agents. Now is the time to return to neuroscience for inspiration.

On the other hand, Andrej Karpathy believes that everyone has an advantage over companies like OpenAI in terms of building AI agents, everyone is currently in a state of equal competition, so he is looking forward to seeing the results in this regard:

AI agents represent a crazy future. Although it may be a bit far away, the AI agents built by everyone present today are already at the forefront of AI agent capabilities. Now all the institutions that are working on large language models, such as OpenAI, etc., I don’t think they are at the forefront of this field, and the forefront is everyone here.

Google's AI team DeepMind's latest paper introduces an AI agent capable of self-improvement - RoboCat, which is essentially a software program powered by AI, which is equivalent to the "brain" of the robot. The robot supported by it is different from traditional robots in that RoboCat is more "universal" and can achieve self-improvement and self-improvement.

Embodied intelligence is more valuable than a humanoid robot

Embodied intelligence is equivalent to the brain of AI, and the carrier of this brain can be in any form. It could be a robotic arm, a robot dog, or even a car.

On the other hand, the reason why humanoid robots are regarded as a not-so-intelligent steel giant at the moment is because of the lack of AI brain + inflexible body.

To put it simply, a large model like GPT-4 has no real impact on the physical world, while embodied intelligence has an extra body, which collects environmental information through sensors, uses mechanical actuators for physical operations, or uses robots to Real-time interaction with humans and the environment with concrete entities such as

Musk once said that although everyone may have a humanoid robot one day in the future, the currently displayed Optimus humanoid robot products can only perform repetitive simple labor.

The goal of embodied intelligence is to enable machines to better understand and adapt to complex environments, solve problems more efficiently, and possess more flexible behavior capabilities. By integrating the processes of perception, decision-making, and execution, embodied intelligence enables machines to be closer to the performance of human intelligence, thus playing an important role in robotics, autonomous driving, and smart manufacturing.

Karpathy said bluntly that 7 years ago, the time to study AI agents was not yet ripe, and the results were not good due to technical limitations, so he and OpenAI changed their direction and began to study large language models.

Now that there are new technical means to study AI agents, the situation is completely different from 2016:

The simplest example is that now no one uses reinforcement learning methods to study AI agents like in 2016. The current research methods and directions were unimaginable back then.

The next wave of AI?

The emergence of large language models has brought new possibilities to the construction of embodied agents. Because LLM-based agents can use the world knowledge contained in the pre-trained model to generate consistent action plans or executable strategies, it is very suitable for tasks such as games and robots.

DeepMind's RoboCat is just one of the leading examples of AI-enabled robots.

Since the beginning of this year, several companies have applied language models to robots: in early 2023, Google launched the visual language model PaLM-E and applied it to industrial robots; in April, Alibaba connected the Qianwen large model to industrial robots; In May, the Tesla humanoid robot Optimus demonstrated precise control and perception capabilities. In the same month, Nvidia released a new autonomous mobile robot platform.

Thanks to this, the robot avatar embodied intelligence blessed by artificial intelligence has attracted widespread attention from all over the world.

Musk stated at Tesla’s 2023 shareholder meeting that humanoid robots will be Tesla’s main source of long-term value in the future:

"If the ratio of humanoid robots to humans is about 2 to 1, then people's demand for robots may be 10 billion or even 20 billion, far exceeding the number of electric vehicles."

Nvidia founder Huang Renxun also said at the ITF World 2023 Semiconductor Conference that the next wave of AI will be "embodied intelligence". Wall Street News previously pointed out that Guosheng Securities analysts believe that embodied intelligence has the characteristics of physical feedback and physical output, and can become a new carrier of communication, computing and storage:

In the future, embodied intelligence will increasingly emphasize the matching and coupling of edge communication capabilities and edge computing power.

The body of AI is actually not the most important thing. The core should be to develop the AI brain and open up human-computer interaction methods, so that AI can actively perceive the physical world, and the anthropomorphic thinking path can achieve the behavioral feedback that humans expect. Machine vision and multi-modal large models are the two keys to open this world.