Following ChatGPT’s introduction of large language models into everyday life in November 2022, the robotics sector has been waiting for a similar turning point. Les Karpas, Global Head of Physical AI at Nvidia Inception, will discuss the obstacles facing robots and potential opportunities during the “Robots Are Waiting for Their ChatGPT Moment. Here Is What Is Standing in the Way” session at TechCrunch Disrupt 2026.
According to Karpas, the main problem facing general-purpose robots is the lack of an internet-scale physical AI dataset like those used by OpenAI and Anthropic for their language models. Even Waymo and its competitors built the data for their self-driving vehicle fleets through years of miles traveled, and these data pools continue to grow in new markets. Startups, meanwhile, are trying to artificially create the required scale through simulation, synthetic data, and foundation models trained simultaneously across different types of robots.
This gap between the digital world and the physical environment will be the main focus of the event’s Real World AI Stage. Karpas will be joined by the founders of Shield AI, Colossal Biosciences, FieldAI, and Foxglove, who will share the opportunities and challenges they face.
At Nvidia, Karpas manages relationships with startups in robotics, automotive, manufacturing, mobility, and smart cities; he previously worked at Stanley Black & Decker, Intellectual Ventures, Herman Miller, iRobot, and Cirque du Soleil. Over the course of his career, he has held roles in architecture, manufacturing engineering, as a startup CEO, as a venture studio executive, and as a corporate venture capitalist.
TechCrunch Disrupt 2026 will host 10,000+ technology leaders, founders, and investors at Moscone West in San Francisco on October 13-15. Discounts of up to $200 are available on tickets before September 25, while group purchases receive discounts of up to 30%. Reservations for exhibition tables close on September 18.
Why it matters
The lack of a robotics equivalent to the vast internet-sourced dataset available for language models highlights a key difference that directly limits the development of general-purpose systems. This is particularly relevant to startups that do not have the same resources as autonomous vehicle companies, which rely on long-term field operations to collect data. Although simulation, synthetic data, and foundation models that train different types of robots together offer options for closing this gap, it remains unclear to what extent they can account for the diversity of real-world physical environments. The discussion’s scope, extending from automotive to manufacturing and smart cities, shows that the data problem is not limited to a single type of robot. However, because the content serves to promote the event and ticket sales, it gives readers no additional basis for determining whether the assessments presented are supported by independent evidence.
Background
ChatGPT is not a new name in the FikirPilot archive: over the past 90 days, we have published 43 news stories mentioning it, the most recent dated September 15, 2026.