
世界“很可能”将进入一个“软件系统前所未有的安全时代”,AMI(先进机器智能)创始人兼董事长杨立昆周二在Semafor举办的“硅谷与世界”活动上表示。
The world is “probably” going to enter an era of “unprecedented safety of software systems,” AMI (Advanced Machine Intelligence) founder and chair Yann LeCun said on Tuesday at Semafor’s Silicon Valley & The World.
“我对AI很少感到担忧,”杨立昆对Semafor的Shelly Banjo说。他预测,会出现一段短暂的时期,“在此期间AI工具会助长网络攻击,使攻击者比防御者更占优势”,但这“只是一个短暂的过渡期”。他提出,未来的软件将通过AI得到“加固”。
“I’m worried about very few things when it comes to AI,” LeCun told Semafor’s Shelly Banjo. LeCun predicts a short period “where AI tools facilitate cyberattacks and favor the attacker over the defender,” but that will be “just a short period of transition.”Future software, he suggested, will be “hardened” by AI.
杨立昆是2018年图灵奖得主,长期以来一直怀疑扩大大型语言模型的规模是否会带来超级智能或通用人工智能。他于2025年离开Meta,创立了自己的世界模型公司AMI,该公司筹集了超过10亿美元的种子资金,是欧洲公司有史以来最大的一笔。他告诉Banjo:“我觉得这么大的时候,它就被称为椰子了。”
LeCun, a winner of the 2018 Turing Award, has long been skeptical that scaling large language models will usher in superintelligence or artificial general intelligence. He left Meta to found his world-model company AMI in 2025, which raised more than $1 billion in seed funding, the largest ever for a European company. He told Banjo that “I think it’s called a coconut when it’s that big.”
他周二谈到了世界模型在物理AI中的实际应用,但承认其发展还有很长的路要走。“我们需要AI领域的一些概念性突破,才能制造出有用的家用机器人,”他说。“目前我们可以在受控环境良好的情况下将人形机器人用于非常狭窄的任务,”但家用机器人“不是明天就能实现的”。他表示,他和其他从事世界模型研究的同行正转向物理AI,因为这是大型语言模型的“完全盲区”。
He spoke on Tuesday about world models’ practical application in physical AI, but acknowledged there is a long way to go in their development. “We’re going to need a few conceptual breakthroughs in AI to make domestic robots that are useful,” he said. “We can use humanoid robots for very narrow tasks in well-controlled environments at the moment,” but domestic robots are “not for tomorrow.”He said he and his peers working on world models are gravitating toward physical AI because it’s a “complete blind spot” of LLMs.
当被问及开源模型时,他表示,“最强大的开源系统是中国人开发的”,这“让美国很多人感到非常难过,包括现政府中的一些人”。
When asked about open source models, he remarked that “it made a lot of people really sad in the US, including people who are in the current government, that the most powerful open source systems are Chinese.”