About this video

• Video Title: OpenAI’s Sam Altman Talks ChatGPT, AI Agents and Superintelligence — Live at TED2025
• Channel: TED
• Speakers: Chris Anderson, Sam Altman
• Duration: 47:30

Overview

This video features a conversation between Chris Anderson and Sam Altman at TED2025, discussing the latest advancements and implications of OpenAI's AI models, including ChatGPT, Sora, and AI agents. They explore topics such as AI's impact on creative industries, the development of AGI, safety concerns, and the future of AI in scientific discovery and daily life.

Key takeaways

• AI Advancements: OpenAI continues to release powerful AI models, with GPT-4o integrating advanced intelligence for tasks like image generation and reasoning, as demonstrated by Sora and its ability to create diagrams distinguishing intelligence from consciousness.
• Impact on Creativity and Economics: The increasing capabilities of AI raise concerns about intellectual property and the future of creative jobs. Altman suggests new economic models will be needed as AI democratizes creativity, while emphasizing that humans will remain central.
• Open Source vs. Proprietary Models: OpenAI plans to release a powerful open-source model, aiming to be at the frontier of current capabilities, acknowledging the important role of open-source AI in the broader ecosystem.
• User Growth and Product Development: ChatGPT has seen unprecedented growth, reaching 500 million weekly active users. OpenAI is focusing on building a great product and integrated experience, not just powerful models, with features like "Memory" aiming to make AI a more personalized companion.
• AI Safety and Ethics: Significant concerns exist regarding AI safety, misuse, and the potential for loss of control, especially with agentic AI. OpenAI is developing frameworks to address these risks, though defining AGI and ensuring safety remain complex challenges.
• Future of AI: Altman expresses excitement about AI for scientific discovery and its potential to accelerate progress against diseases and in physics. He also anticipates significant advancements in agentic software engineering in the coming months.
• AGI and Superintelligence: While acknowledging the ongoing development towards AGI, Altman emphasizes the exponential progress of AI and the need to focus on building safe systems that can adapt to increasingly capable AI, rather than fixating on a specific definition of AGI.
• The Role of Agency: Agentic AI, capable of pursuing projects independently, presents both immense opportunities and significant safety challenges. OpenAI is developing guardrails, recognizing that user trust and safety are paramount for adoption.
• Governance and Societal Impact: The discussion touches on the need for external safety testing for advanced models and exploring new models for collective decision-making on AI's future, potentially using AI itself to understand societal preferences.
• Personal Reflections and Future Vision: Altman reflects on the personal impact of fatherhood and his belief that AI will lead to a world of material abundance and increased human potential, though cautioning against unmanaged risks.

Sam Altman explains that AGI is not simply about being able to answer any question intelligently, as current models like ChatGPT can do. He states that AGI does not continuously learn and improve on its own, nor can it independently go and get better at something it is currently weak at, or discover new science to update its understanding. He also notes that AGI, in his view, would be able to perform any knowledge work a human could do in front of a computer and perform tasks like navigating the internet, calling people, and accessing files autonomously, which current systems cannot do. He acknowledges that even defining AGI is challenging, with different researchers having various definitions.