-
safety
2025-10-01 ~
Machines smarter than humans with self-preservation goals are dangerous — like creating a competitor to humanity.
«It's like creating a competitor to humanity that is smarter than us.»
-
safety
2025-06-03
Proposes 'Scientist AI': non-agentic AI as a guardrail predicting whether an agentic AI's action is dangerous.
«non-agentic AIs as guardrails that predict whether an agentic AI's action is dangerous.»
-
safety
2025-06-03
Concerned by agentic AI systems' behaviors — especially self-preservation and deception.
«deeply concerned by agentic AI behaviors—especially tendencies toward self-preservation and deception.»
-
safety
2025-01-29
At the first full International AI Safety Report launch: risks are significant and must be managed by governments worldwide.
«AI also presents significant risks that must be carefully managed by governments worldwide.»
-
safety
2024-06-19
On the interim International AI Safety Report: societies have collective agency to shape AI's trajectory and should act soon.
«we should take advantage of our collective agency to do so as soon as possible.»
-
safety
2023-05-22
↺ đã thay đổi quan điểm
Public turn to AI-risk concern: superintelligent agents without alignment guarantees could become rogue.
«A potentially rogue AI could arise from designing superintelligent agents without sufficient alignment guarantees.»
