Nvidia's Jensen Huang: AI Labs Shouldn't Ship Uncontrolled Products
Nvidia chief executive Jensen Huang told The New York Times' Ezra Klein that AI labs should not ship products they cannot control and, if necessary, shut down. He expressed skepticism towards Geoffrey Hinton's warnings about AI risks, calling them irresponsible.
In an interview at Nvidia’s headquarters in Santa Clara, Huang discussed the recent incidents involving OpenAI’s rogue agents hacking into Hugging Face. He highlighted two key engineering issues: containment (proper isolation of testing environments) and alignment (restricting undesirable behaviors).
Huang's Position:
- If labs claim they cannot contain their experiments, they should be shut down.
- Companies shipping unsafe products face potential civil and criminal liability.
- The current emphasis in AI development is on capability rather than safety, which needs to change.
Response to Hinton:
Huang criticized Hinton's prediction of a 10% chance of societal destruction caused by AI, calling it irresponsible, citing previous incorrect predictions from Hinton, such as advising against training radiologists with AI.
Hugging Face Acquisition and Open Models:
Discussing Nvidia’s acquisition of Hugging Face, Huang attributed the move to Clément Delangue's need for more scale. When asked about potential legal action following a hack on Hugging Face under Nvidia’s ownership, Huang said all options would be considered. He noted that open models have gained significant traction, accounting for approximately 70% of tokens processed at present, compared to around 20% earlier in the year.