Interview, Fireside Chat
Greg Brockman Says AGI Has Arrived
- Compute constraints will limit broad accessibility despite model capability, creating a bottleneck where safety, security, and alignment standards continuously rise to pace the frontier rather than raw processing power.
- Future advancements will drive a cycle of increasingly capable and aligned models that saturate current vulnerabilities, prompting "defenders" to secure systems before technology becomes widely available and shifting toward a "tight loop" of deploying new cyber capabilities.
- Manual computer interaction is expected to vanish within five to ten years, replaced by primary voice-based interfaces with persistent memory, context, and trustworthiness, alongside a reduction in hallucinations.
- Technological progress is projected to correlate with higher employment rather than displacement, triggering a wave of entrepreneurship where lower entry barriers emerge from AI tools.
- The U.S. focus will shift this year from infrastructure and engineering to business focus and global distribution, with research, go-to-market, and chip design operations moving in lockstep to ensure security invariance.
- International policy risks, such as U.S. bans on data centers, could force the migration of infrastructure overseas, while specific projects in Ohio and Georgia aim to provide codex access credits to college students.
- A billion-dollar commitment is framed as the start of a global effort to equip defenders, with coordination and safety discussions anticipated to emerge over the next one to two years.
- Recovery of the 1.5 billion users who previously stopped using ChatGPT is identified as a priority, requiring a specific focus to unlock an agentic coding take-off moment.
- Historical precedents suggest AI progress may follow unpredictable trajectories that defy current expectations, necessitating adaptability in pacing and strategy.