Sam Altman says he expected GPT-4 to disrupt software and business much faster than it did. The missing factor was economic inertia: people keep buying familiar products, using established tools and defining productive work through habits that persist even when a better technical option is available.
Sam Altman uses his own limited adoption of Codex as an example. Although an agent could handle emailAn AI agent is a system that observes context, decides what to do, and takes actions through tools to pursue a goal., task lists and routine computer work, moving between familiar interfaces still feels like workAI adoption friction is the effort, uncertainty, disruption, or risk that makes people delay or avoid moving an AI capability into routine use.. He argues that AI products have not yet made the new workflow seamless enough to overcome those ingrained behaviors.
The video separates slow social adoption from rapid capability growthThe AI capability-adoption gap is the difference between what current AI systems can technically do and what people or organizations routinely use them to do.. Sam Altman expects safety decisions to become harder as models surpass the smartest people, and points to cases where OpenAI paused some work to strengthen cybersecurity controlsAn AI cybersecurity safeguard is a control that protects AI models, data, tools, infrastructure, and users from unauthorized access, manipulation, or abuse. before continuing development.
Sam Altman also acknowledges that the AI industry damaged public trust by warning about catastrophic risks and job losses while racing ahead. His proposed product direction is a persistent, context-aware interfaceA context-aware AI interface preserves and retrieves relevant user, task, and application state so assistance can continue without repeated setup. and an API that make intelligence broadly useful without requiring OpenAI to build every downstream product.
Watch on YouTube
