LeCun opposes slowing AI and backs stronger cyber defence
Yann LeCun opposed calls to slow AI development at an industrial cooperation briefing in Seoul. The AMI Labs founder argued that progress would be gradual and that cyber defenders could draw on substantial resources. He also acknowledged difficulties in guaranteeing safe language-model answers and described systems that predict the consequences of actions. His remarks present a proposed approach, not a new safety result.
Artificial Intelligence··Night
The Seoul briefing puts gradual progress against slowdown calls
Yann LeCun, a New York University professor and founder of AI developer AMI Labs, opposed slowing AI development at an industrial-cooperation press briefing in Seoul on October 12. He argued that progress would come gradually from many innovations around the world. His position rejects predictions of a single sudden leap in which machines surpass human capabilities across the board.[1], [2]
The meeting at the Westin Josun hotel concerned international cooperation on industrial AI. LeCun spoke as founder and chairman of AMI Labs, following his earlier work as Meta’s chief AI scientist. He is also a recipient of the Turing Award. The intervention sets out his view of the technology’s direction and safety, rather than presenting results from a newly released model.[1]
Cyber defence becomes his answer to recent attacks
LeCun discussed recent cyber attacks on South Korean financial institutions and attributed the problems to weak defences, limited experience and carelessness. That explanation is his assessment, rather than an independently investigated account of each incident. He advocated strengthening defensive capabilities as the response, while arguing against holding back development of the technology.[1], [2]
In his account, defenders can draw on more people, highly educated specialists, funding and wider knowledge. He also described the ability to work openly as an advantage over attackers. News1 reports his argument that open communities give defensive specialists access to AI tools. Those proposed advantages do not establish that every attack has been prevented.[1], [2]
World models predict actions while safeguards remain difficult
LeCun acknowledged that language-model safeguards can be bypassed and that a safe answer cannot easily be guaranteed in every situation. He described objective-driven systems that anticipate the consequences of actions and operate within constraints. The approach connects a proposed goal with predictions about possible actions, rather than relying only on generating a response.[1]
For robotics, he contrasted language models’ mathematics and coding abilities with the continuing difficulty of ordinary physical tasks. World models, as he described them, predict what follows from an action and plan steps toward a goal. He pointed to Korea’s manufacturing base and researchers as potential strengths and advocated improving the university and research environment to attract talent.[2]
