Openai raises robots in the race towards AGI

Renewable focus on robots may indicate that Openai believes that access to artificial general intelligence (AGI) – which goes beyond human intelligence – requires development algorithms capable of interacting with the material world.

Openai not noticeable robots research in its early years, including developing an algorithm capable of solving a Rubik cube using human hand In 2019, the company Robotat efforts were closed in 2021However, to focus on algorithms, including large language models that paid modern breakthroughs such as Chatgpt. Openai Restart the work on robots last yearAnd Information received in December 2024 The company was considering developing its human robots.

Stephanie Tellex, a robot scientist at the University of Brown, says that building more effective robots will include the design and training of artificial intelligence models capable of “handling high-dimensional perception inputs, and producing high-dimensional high-dimensional material outputs”-models that can see and behave with high sincerity. Tellex is not aware of Openai’s plans specifically.

Although there are already pioneering models in conversation, thinking, coding and generating images and video, Openai will run a series of strong competitors as they seek to develop algorithms for more capable human robots. A handful of human startup, including appearanceand LightnessAnd ApptronikIt has appeared over the past few years, and some major AI companies, including Tesla and Google, have invested in the development and test of Humanoids. “I don’t see them enjoying any magical advantage over anyone else,” says Tellex says.

Mankind has become increasingly popular Since the devices and programs needed to build the initial models become more common. Although Humanoids are still expensive and difficult to develop, new types of engines and other components have made the cheapest and easiest to collect operating systems. Programs like Nafidia Isaac The robot development platform has made it easy to write the code needed to control and train human systems.

Humanoid Hype is also a building. Invested the owners of investment capital More than 5 billion dollars In human startups since the beginning of 2024. Morgan Stanley believes that the human industry It can be equal to $ 5 trillion by 2050.

While Humanoids can perform impressive exploits such as dance, it still lacks the intelligence required to work in complex and unexpected environments, or “unorganized”. To get this, they will need algorithms that go beyond understanding the great language model for the material world. These systems should be able to control the limbs and pesticides in order to walk and treat material elements. Some research groups have started to show progress in development In general, artificial intelligence models For robots.

At the same time, it has become increasingly clear that new ideas are needed to push artificial intelligence forward. The last disappointment from Openai’s GPT-5 It is part of a broader awareness that access to human intelligence will require new ways of research.

“It was erected on GPT-5,” says Tellex says. “They need to move towards the material world.”