DeepMind co-founder warns AI safety can't lag behind progress
Shane Legg joins the debate over whether AI companies should slow down to prioritize safety over raw capability advances.
In brief
Shane Legg joins the debate over whether AI companies should slow down to prioritize safety over raw capability advances. Leading AI researchers are increasingly arguing that safety must keep pace with capability advances, though they disagree on whether slowing development is the right approach. Originally reported…
DeepMind co-founder warns on AI safety
Shane Legg says AI capabilities are advancing too fast and safety must keep pace. He launched the DeepMind Institute to study the implications of artificial general intelligence.
“We can't let capabilities get ahead of safety.”
Anthropic CEO proposes slowing AI releases
Dario Amodei published an essay titled 'We Must Pace the Frontier' arguing companies should slow new AI model releases to address safety concerns. The proposal sparked debate across Silicon Valley.
“We Must Pace the Frontier”
Legg cautious on pacing proposal
While Legg called Amodei's essay 'interesting directionally,' he said the details of how pacing would work in practice need more consideration. He remains comfortable with his 50% chance prediction for minimal AGI by 2028.
“we need to really work through the details of that and how that would work in practice.”
Legg skeptical AGI has arrived
Despite recent claims from OpenAI and Nvidia executives, Legg said it's too early to declare AGI has arrived. He maintains his long-standing view on the timeline for artificial general intelligence development.
Zuckerberg: alignment is competitive advantage
Meta CEO Mark Zuckerberg argues that trust and alignment will become the key differentiators between AI models. Labs that ignore alignment will fall behind competitors, he said.
“trust and alignment are quickly becoming the most important capabilities that will differentiate agents and models.”
Meta delayed Muse for safety reasons
Meta postponed releasing its Muse AI for several months to focus on safety and security. Zuckerberg said the company made this decision independently without calling for industry-wide delays.
“We just did it as part of our day-to-day work because it was clearly the right thing for people and for us.”
Liability creates safety incentives
Zuckerberg noted that labs face significant liability if their models cause harm, giving them strong incentive to prevent problems. He also advocated for independent evaluators and diverse evaluation ecosystems.
“Labs face significant liability if their models cause harm, so they have a strong incentive to prevent this as well.”
Debate continues on best approach
While industry leaders agree safety matters, they disagree on whether mandatory pacing is necessary. Some believe market forces and liability will naturally drive safety improvements.
The one thing
Leading AI researchers are increasingly arguing that safety must keep pace with capability advances, though they disagree on whether slowing development is the right approach.
