LinkedIn联合创始人Reid Hoffman利用人工智能技术，包括GPT模型和语音合成工具，创建了一个AI版的自己，被称为“Reid AI”。这个AI双胞胎不仅能够模仿他的声音，还能用包括中文、日文、印地语和意大利语在内的多种语言发表演讲，极大地拓展了信息的传播范围和跨文化交流的可能性。Hoffman将此视为对深度伪造技术潜力的探索，旨在证明这项技术在透明和负责任的使用下，可以促进连接而非传播虚假信息。尽管目前AI双胞胎仅用于私下实验和翻译，但他强调了技术向善应用的必要性，同时也警惕了被滥用的风险。

🌐 **AI双胞胎拓展信息传播与跨文化交流**：Reid Hoffman构建了一个AI版本的自己，能够模仿其声音并使用多种语言（包括中文、日文、印地语和意大利语）进行演讲。这显著提升了信息传播的触达范围，并为跨文化交流提供了新的可能性，让原本 Hoffman 无法触及的语言受众也能接收到信息。

💡 **探索深度伪造技术的善用与伦理边界**：Hoffman将这项技术称为“深度伪造”，并致力于探索其积极的应用方式，以对抗其常与虚假信息传播联系的负面印象。他通过创建AI双胞胎，旨在证明在透明和谨慎使用的前提下，深度伪造技术可以用于促进人与人之间的连接。

⚠️ **谨慎对待AI双胞胎的潜在风险**：尽管AI双胞胎带来了诸多益处，Hoffman也承认其局限性，特别是被黑客利用来传播不当言论的风险。因此，他目前将AI双胞胎限制在私下实验和受控环境中，并未向公众广泛开放，以防范潜在的滥用。

✅ **透明与负责任是AI技术应用的关键**：Hoffman强调，在开发和使用AI双胞胎时，透明度至关重要。AI的合成性质需要被明确告知，其使用也应受到严格的控制。他认为，通过道德设计和有意识的运用，可以引导这项技术朝着有益于社会的方向发展。

Reid Hoffman built an AI version of himself to explore ethical uses of deepfake tech.His AI twin delivers speeches in multiple languages using his cloned voice.Hoffman said deepfakes can connect people — if used with transparency and care.

Reid Hoffman, the LinkedIn cofounder and veteran Silicon Valley investor, has created an AI-powered version of himself — and the results are both eye-opening and a little surreal.

In a Monday interview on the "American Optimist" podcast, Hoffman revealed he trained a "deepfake twin" using a custom GPT model, ElevenLabs for voice synthesis, and video-generation tools to explore whether a technology widely associated with disinformation could also be repurposed for good.

"You have this technology that most people call deepfake, and deepfake just sounds like it's bad," Hoffman said. "So I was like, let's start experimenting with it and see how it can work in better ways."

The AI clone, which he calls "Reid AI," is designed to act and sound like him — and, in some cases, speak languages Hoffman doesn't.

After delivering a speech at the University of Perugia in Italy in May 2024, he used his AI twin to deliver the same message in Hindi, Chinese, Japanese, and Italian.

"I've never heard my voice speaking Hindi; I've never heard my voice speaking Japanese," he said. "And it was like, wow, it is clearly my voice."

Hoffman described the projects as "raw experimentation" with surprising upsides, including increasing accessibility and expanding the reach of ideas.

But he also acknowledged serious limitations.

"I haven't put it out in the wild," he said, citing the risk of hackers making the AI say "crazy stuff."

Deepfakes have already been misused in several ways. These include impersonating former President Joe Biden in AI-generated robocalls urging voters not to participate ahead of the 2024 New Hampshire primary and spreading fake surrender messages from Ukrainian President Volodymyr Zelenskyy.

For now, Hoffman's clone remains private and serves as a tool for translation, experimentation, and testing new ways to communicate across cultures and platforms.

Hoffman built the projects with transparency in mind: the AI's synthetic nature is disclosed, and its use remains tightly controlled.

His takeaway is that not all deepfakes are inherently harmful. With ethical design and intentional use, the same technology driving disinformation could help people connect across languages and cultures.

"What we're trying to do, as technooptimists say, is how do we shape it so it's applied well," he said.

Hoffman isn't the only one to have experimented with AI clones.

"The Diary of a CEO" host Steven Bartlett launched a new podcast in May, "100 CEOs with Steven Bartlett," using an AI-generated version of himself to narrate stories about business leaders.