Greg Brockman's OG Blog (blog.gregbrockman.com)

Greg Brockman argues in his OG Blog that AI has passed a “utility threshold”: models like GPT-3, Codex and DALL·E 2 are no longer just research curiosities but practical tools that perform tasks no other software can. Building these models is framed as an exploratory frontier—producing them uncovers unknown capabilities, advances science, and unlocks novel product applications. Crucially, because modern AI is fundamentally about creating and operating complex software systems, top engineers can contribute at the same level as top researchers to future progress. The piece reframes career paths and priorities for the AI/ML community: emphasis shifts from purely theoretical research to ML engineering—systems design, productionizing models, scaling training and inference, tooling, evaluation and safety, and integrating emergent behaviors into products. That means practical engineering skills (large-scale distributed systems, robust deployment pipelines, metrics and safety frameworks) are now core to making and discovering the next generation of capabilities. For practitioners and teams, Brockman’s message is a call to action: treat model development as a systems problem, invest in engineering rigor, and expect engineers and researchers to collaborate tightly to explore and harness new, sometimes unpredictable, model behaviors.