Thinking Machines 谈 AI 的下一章

Thinking Machines on AI’s next chapter

米拉·穆拉蒂 Mira Murati · 彭博 · 2026-06-04 · 约 28 分钟 · 原视频 ↗

本期速览 · Overview

米拉·穆拉蒂谈她的新实验室，以及构建可理解的 AI。

Mira Murati on her new lab and building understandable AI.

AI 应实现实时、高带宽的协作，而非回合制交互。
AI should enable real-time, high-bandwidth collaboration, not turn-based interaction.
让人类参与循环是使 AI 与人类意图和价值观对齐的关键。
Keeping humans in the loop is key to aligning AI with human intent and values.
我们有能力引导 AI 发展走向协作，而非乌托邦或反乌托邦。
We have agency to steer AI development toward collaboration, not dystopia or utopia.

交互模型实现实时、高带宽的人机协作，而非回合制。
Interaction models enable real-time, high-bandwidth human-AI collaboration, not turn-based.
最先进的 AI 系统应成为思维工具，扩展我们思考的内容。
The most advanced AI systems should be tools for thought, expanding what we think about.
在 AI 进步过程中，让人类参与其中对于对齐意图和价值观至关重要。
Keeping humans in the loop is crucial for aligning intent and values as AI advances.
制度设计、透明度和制衡比个人品格更重要。
Institutional design, transparency, and checks and balances matter more than individual character.
未来并非注定；我们有能力引导 AI 发展走向协作。
The future is not predestined; we have agency to steer AI development toward collaboration.
前沿 AI 的快速进步带来波动，但长期可持续进展是关键。
Compressing progress in frontier AI leads to volatility, but long-term durable progress is key.

交互模型应基于时间而非轮次，以捕捉人类细微差别。
Interaction models should be time-based, not turn-based, to capture human nuance.
ChatGPT 最重要的影响是将 AI 带入公众意识，而不仅是技术。
The most important impact of ChatGPT was bringing AI into public consciousness, not just technology.
讨论过于关注个人品格，而对治理和透明度的关注不足。
The conversation focuses too much on individual character, not enough on governance and transparency.
预测乌托邦或反乌托邦过于简单化；我们在如何构建 AI 上有能动性。
Predicting utopia or dystopia is oversimplified; we have agency in how we build AI.
持久 AI 进步的道路是尽可能让人类参与循环，而非移除他们。
The path to durable AI progress is keeping humans in the loop as long as possible, not removing them.
将进展压缩到短时间内会导致好与坏的结果在几个月内出现。
Compressing progress into short timeframes leads to both good and bad outcomes in months.