OpenAI 的 Sora 团队讨论意外的病毒式成功、将其打造成社交体验的产品决策，以及包括新颖物理发现在内的未来里程碑。
OpenAI's Sora team discusses the unexpected viral success, product decisions behind making it a social experience, and future milestones including novel physics discoveries.
要点 · TL;DR
Sora 的 Cameo 功能让 AI 视频变得个性化，推动创作而非被动消费。 Sora's Cameo feature makes AI video personal, driving creativity over passive consumption.
视频模型将在 2028 年前实现物理模拟的科学突破。 Video models will enable scientific breakthroughs in physics simulation by 2028.
缩放定律适用于视频模型，成本快速降低且智能提升。 Scaling laws apply to video models, with rapid cost reduction and intelligence gains.
核心观点 · Key points
Sora 的社交功能如 Cameo 让 AI 视频变得个性化和人性化，推动创造力而非消费。 Sora's social features like Cameo make AI video personal and human, driving creativity over consumption.
视频模型将在 2028 年前实现物理学模拟的科学突破。 Video models will enable scientific breakthroughs in physics simulation by 2028.
缩放定律适用于视频模型，成本快速降低且智能显著提升。 Scaling laws apply to video models, with rapid cost reduction and intelligence gains.
人类的创作意图对于生成引人入胜的 AI 内容至关重要。 Human creative intent is essential for compelling AI-generated content.
Sora 的产品设计优先赋能创作者和社区，而非被动消费。 Sora's product design prioritizes creator empowerment and community over passive consumption.
反共识 · Contrarian takes
AI 视频模型可以促进更深层次的社交联系，而非孤立。 AI video models can foster deeper social connections rather than isolation.
没有人类策展的纯 AI 生成内容感觉空洞且无趣。 Pure AI-generated feeds without human curation feel hollow and uninteresting.
最大的挑战是消费者采用，而非模型能力。 The biggest challenge is consumer adoption, not model capability.
视频模型通过预测学习世界模型，类似于语言模型。 Video models learn world models through prediction, similar to language models.
变现应使创作者、版权方和 OpenAI 平等受益。 Monetization should benefit creators, rights holders, and OpenAI equally.
本期章节 · Chapters（共 37）
引言与Sora发布Introduction and Sora's Launch
Sora独立应用Sora as an Independent App
最古怪原型：ChatGPT内社交流The wackiest prototype: a social media stream inside ChatGPT