The debate around AI safety has grown considerably louder as models have become more capable and AI agents are given more autonomy for recursive self-improvement, accessing networks and data, and taking actions on behalf of users. At the extreme end of that debate are warnings that advanced AI could eventually become difficult or even impossible for humans to control, ultimately culminating in an existential threat.
NVIDIA CEO Jensen Huang has pushed back on some of those AI apocalypse scenarios, arguing that keeping AI under control ultimately comes down to an engineering problem that NVIDIA is now throwing serious resources behind as well.