“GPT-4 successor models will be shown to have fundamental reliability limitations preventing deployment in high-stakes medical settings without heavy human oversight.”

LLMs have demonstrated significant hallucination and reliability issues limiting autonomous deployment in high-stakes domains. Medical and legal deployments exist but require substantial human oversight, supporting Marcus's core concern while falling short of a blanket prohibition.

LLMs have demonstrated significant hallucination and reliability issues limiting autonomous deployment in high-stakes domains. Medical and legal deployments exist but require substantial human oversight, supporting Marcus's core concern while falling short of a blanket prohibition.