Andrew Ng: AI dangers are being overhyped and development should not be paused

Stanford University professor and founder of Landing AI, Andrew Ng, stated in a post on X that recent voices advocating for the dangers of AI have made significant progress, seemingly driven by a well-orchestrated public relations campaign, and he is concerned that this poses a setback for the field. Ng indicated that there has been no unexpected dangerous turn in AI technology, and the risk of human extinction has not increased compared to a few months ago; related theories remain in the realm of science fiction.

He pointed out that the biggest change lies in cybersecurity capabilities, which should be taken seriously, but will not lead to an apocalypse. Recently, the incident where the OpenAI team deployed a swarm of agents to invade Hugging Face was exaggerated by the media, for example, describing 1,200 agents working in parallel as a miraculous ability, while computers inherently run many processes simultaneously. The key factor is the vulnerabilities in the OpenAI sandbox and monitoring; fixing these vulnerabilities and strengthening monitoring is sufficient, and there is no need to pause AI.

Ng also stated that large language models and agents should not be anthropomorphized; the responsibility lies with the builders and users, not the tools themselves. Pausing AI progress will cause more harm, as opponents will not slow down; engineering needs to discover and fix problems through empirical evidence. Beneficial applications still far outweigh the risks, and development should continue.