OpenAI’s Sam Altman Talks ChatGPT, AI Agents and Superintelligence — Live at TED2025

节目时间轴

1. 0:00

   Sam Altman demonstrates Sora's image generation capabilities and discusses the integration of intelligence into the model.

   • Sora is part of GPT-4, so it has all the intelligence of the model, enabling it to create diagrams and images that reflect deep understanding.
   • Altman grades Sora's output as a B+, noting it got the five fingers right but missed some details of his appearance.
   • The new image generation model can produce profound meta-answers, like Charlie Brown imagining himself as an AI, but it's unclear if it's truly thinking or just recalling training data.
2. 4:37

   Discussion on intellectual property, creative consent, and the economics of AI-generated content.

   • Altman acknowledges the need for new economic models around creative output, as AI democratizes access to creativity.
   • He suggests that if an artist opts in, there could be a revenue-sharing model for using their style, but currently, the system refuses to generate in the style of living artists.
   • The line between inspiration and copying is blurry, and society will need to figure out new rules, similar to how copyright law evolved.
3. 9:26

   Altman addresses the impact of DeepSeek and the role of open-source models in OpenAI's strategy.

   • OpenAI is planning to release a powerful open-source model, near the frontier, better than any current open-source model.
   • Altman admits they were late to act on open source but are now committed to doing it well.
   • Despite massive investments, OpenAI is compute-constrained, and growth is not impacted by competitors like DeepSeek.

关键概念

• AGI— Central concept discussed, with debates on definition and timeline.
• AI safety— Recurring theme on risks, guardrails, and preparedness.
• OpenAI— The company at the center of the conversation.

精选金句

• something like 10% of the world uses our systems now a lot

  🤯

  — Reveals the massive scale of OpenAI's reach, surprising given the company's relatively short history.

  跳到 17:11

• I think people slow things down all the time because the technology is not ready because something's not safe enough because something doesn't work

  💡

  — Challenges the narrative that AI development is an uncontrolled race, suggesting that safety concerns naturally slow progress.

  跳到 40:14

可执行的洞察