The renewed focus on robots will suggest that Openai believes that reaching artificial intelligence (AGI) (more than human intelligence) may require the development of algorithms that can interact with the physical world.
Openai conducted significant robot research in the early years, including the development of an algorithm that can solve one Rubik’s Cube with Human Hands 2019. company Robot jobs closed in 2021However, focusing on algorithms, including large language models that drive recent breakthroughs, e.g. chatgpt. Openai Restarting the robot’s work last yearand Information on December 2024 report The company is considering developing its own humanoid robot.
Stefanie Tellex, a robotist at Brown University, said building more efficient robots would involve designing and training AI models that can “process high-rate rates, high-dimensional perceive inputs, and produce high-rate, high-dimensional physical outputs” – models that can be seen and have high inertia. However, Tellex is not particularly familiar with OpenAI’s plans.
Despite already having industry-leading models for dialogue, reasoning, coding, and image and video generation, Openai will still compete among a range of strong competitors as it tries to develop algorithms for more capable humanoid robots. A few types of human startups, including number,,,,, agileand Apptronikwhich has emerged over the past few years, has also invested in developing and testing humanoid animals. “I can’t see that they have any magical advantages over anyone else,” Tellex said.
Humanoid animals are becoming more and more popular As the hardware and software required to build functional prototypes becomes increasingly common. Although humanoids are still expensive and difficult to develop, new motors and other components make them cheaper and easier to combine functional systems. Such as NVIDIA’s ISAAC The robot development platform also makes it easier to write the code needed to control and train humanoid systems.
Humanoid hype is also being established. Venture capitalists have invested More than $5 billion Among humanoid startups since early 2024 By 2050, it may be worth $5 trillion.
Although humanoids can perform impressive feats such as dancing, they still lack the intelligence they need to operate in complex and unpredictable or “unstructured” environments. To get this, they will need algorithms that go beyond the understanding of the physical world of large language models. These systems must be able to control limbs and handshakes in order to walk and manipulate physical objects. Some research groups began to show progress in development More common AI models For robots.
At the same time, it is becoming increasingly obvious that new ideas may be needed to push AI forward. Recent disappointment Openai’s GPT-5 is part of a broader understanding that achieving human intelligence will require new research paths.
“They’ve gotten asymptotic to the GPT-5,” Tellex said. “They need to move towards the physical world.”