OpenAI 如何取胜，以及 ChatGPT 的未来

How OpenAI wins, and ChatGPT’s future

萨姆·奥尔特曼 Sam Altman · Big Technology · 2025-12-18 · 约 58 分钟 · 原视频 ↗

本期速览 · Overview

奥尔特曼谈竞争、基建，以及 ChatGPT 的下一步。

Altman on competition, the buildout, and where ChatGPT goes next.

前沿模型将创造大部分经济价值，而商品化模型服务于日常使用。
Frontier models will create most economic value while commoditized models serve everyday use.
为 AI 优先世界从头设计产品优于将 AI 附加到现有产品上。
Redesigning products from scratch for AI-first world beats bolting AI onto existing ones.
AI 记忆将变得超人类，记住用户生活的每一个细节。
AI memory will become superhuman, remembering every detail of a user's life.

日常使用的模型会商品化，但前沿模型将创造最大的经济价值。
Models will commoditize for everyday use, but frontier models will create most economic value.
将 AI 附加到现有产品上不如为 AI 优先的世界从头重新设计。
Bolting AI onto existing products is inferior to redesigning them from scratch for an AI-first world.
AI 记忆将超越人类，记住用户生活的每个细节和偏好。
AI memory will become superhuman, remembering every detail of a user's life and preferences.
当前模型（如 GPT-5.2）未开发的经济价值巨大且被低估。
The overhang of untapped economic value from current models like GPT-5.2 is massive and underestimated.
科学发现是算力最高杠杆的用途，AI 将显著加速它。
Scientific discovery is the highest-leverage use of compute, and AI will accelerate it significantly.
AGI 定义不清，可能已经实现；超级智能是更清晰的未来目标。
AGI is underdefined and may have already been achieved; superintelligence is a clearer future goal.

将 AI 嫁接到现有产品上不如从头重新设计效果好。
Bolting AI onto existing products won't work as well as redesigning from scratch.
模型能力的过剩非常巨大；大多数人未充分利用现有模型。
The overhang of model capability is massive; most people underuse current models.
AGI 定义不明确，可能已经悄然过去而没有明确的里程碑。
AGI is underdefined and may have already passed us by without a clear milestone.
由人类治理的 AI CEO 可能成为未来合理的结构。
An AI CEO governed by humans could be a reasonable future structure.
如果今天算力翻倍，我们的收入也会翻倍。
We would be double the revenue if we had double the compute today.
当前的聊天界面出人意料地有粘性；我低估了它的力量。
The current chat interface is surprisingly sticky; I underestimated its power.