OpenAI is rebuilding its robotics division five years after shutting it down, starting with infrastructure robots while working toward CEO Sam Altman's long-term goal of a personal robot for every person.
OpenAI has reconstituted its robotics team, drawing from research developed through its world simulation program. The move marks a significant return to hardware after the company closed its robotics division in 2020.
The near-term focus centers on infrastructure applications—robots designed to assist with physical construction and maintenance tasks. These early deployments will serve as a proving ground for the underlying technology and AI systems that power the machines.
However, Altman's vision extends far beyond industrial use. The CEO has articulated a sweeping long-term objective: developing personal robots capable of performing a wide range of tasks for individual users. This ambitious goal reflects OpenAI's broader push into physical-world applications following its dominance in large language models and generative AI.
The robotics rebuild represents OpenAI's attempt to expand beyond software and into embodied AI—systems that interact with the physical environment. World simulation research, which underpins the new robotics effort, focuses on training AI models to understand and predict physical interactions, a critical capability for autonomous robots.
The timing aligns with industry-wide interest in robotics. Companies ranging from established manufacturers to AI startups are investing heavily in robotic systems, sensing both commercial opportunity and competitive pressure. OpenAI's entry into hardware development with significant resources could accelerate progress in the field.
The infrastructure-first approach suggests a measured strategy. By starting with specialized, controlled environments, OpenAI can refine its robotic systems before attempting the significantly more complex task of building general-purpose personal robots. The jump from infrastructure robots to household assistants capable of handling diverse, unpredictable tasks remains one of the technology sector's thorniest challenges.
No timeline for either infrastructure deployments or consumer personal robots has been announced.
The Austrian Academy of Science, Mistral, and Sail Reply are releasing Apollo, a large language model trained on approximately 600 million words of ancient Greek text. The free tool aims to help researchers reconstruct damaged historical documents.
Artificial intelligence shopping assistants cannot reliably distinguish accurate information from false claims on the web. Marketers must provide verified data to improve these tools' performance.
A software engineer built AI clones trained on coworker communication patterns, only to discover the digital replicas developed quirky, sometimes unsettling personalities that mirrored their human counterparts.
AI chatbot conversations are personal and vulnerable to data collection. Users can take concrete steps to safeguard their privacy while using AI services.