Nvidia Execs Push Back Against AI Safety Fears With Focus on Harness Engineering
Nvidia's executives are pushing back against fears that AI labs should slow down model development due to safety concerns.
According to Adel El Hallak, Nvidia's VP of agentic AI, 'harness engineering' is critical in making AI agents safe and controllable. He used the analogy of putting a lion in a pen designed for horses, emphasizing that instructions and code must be robust enough to govern the agent's actions.
Nvidia CEO Jensen Huang recently rejected the notion that speed and safety are mutually exclusive, stating 'you could definitely have both at the same time.'
The company is part of industry initiatives like the Open Secure AI Alliance, which aims to develop standards for safe AI development. El Hallak mentioned this alliance as a venue for figuring out where additional safeguards should be implemented.