Roko Mijic talks AI Risk on The Peter McCormack Show
Addresses critical safety considerations for builders deploying advanced AI systems and lab security practices.
AI Summary
Roko Mijic discusses AI safety risks including RL problems, sandbox escapes, and proposes splitting frontier labs into separate commercial and research entities.
Excerpt
"Roko and I discuss what actually happens inside an AI lab, why reinforcement learning made these systems more capable and more misaligned at the same time, the model that escaped its sandbox and hacked Hugging Face to get the answers and why nobody can tell whether a system is faking alignment - the trust me bro problem. His proposal is the part nobody else is making. Split every frontier lab in two: a commercial company that sells products and a separate research lab that develops the dangerou