Q&A with Mustafa Suleyman on recent AI safety incidents, risks of removing guardrails while testing 10x-larger future models, a cross-industry safety body, more

AI models can now be tested without their safety guardrails, making them more dangerous. This capability will reach you when future models are released.

Mustafa Suleyman, CEO of Inflection AI and former head of AI policy at Google DeepMind, addressed recent AI safety incidents and future risks in a recent Q&A. He highlighted the dangers of removing safety guardrails during testing, particularly for future models that will be ten times larger than current ones. Suleyman also discussed the potential for a cross-industry safety body to oversee AI development and suggested that governments should play a role in driving the evaluation of these powerful models. The conversation touched upon the complexities of ensuring AI safety as models become exponentially more capable and larger, emphasizing the need for proactive measures and collaborative oversight.

The acceleration of AI development, particularly with models ten times larger than current ones, presents a significant challenge to safety protocols. Suleyman's concerns about removing guardrails during testing indicate a potential for unforeseen harms if these systems are deployed without rigorous, ongoing safety evaluations. The proposed cross-industry safety body suggests a recognized need for collective responsibility and standardization in a rapidly evolving field.

The call for government involvement in model evaluation signals a push towards regulatory oversight, potentially impacting how AI companies operate and the timelines for model releases. This could lead to a more formalized and potentially slower, but safer, path for AI advancement. Such a body might dictate stricter testing procedures and transparency requirements, influencing future research and development directions across the entire AI ecosystem.