What happened in this episode?
OpenAI president Greg Brockman tells the No Priors podcast he believes AI systems can now generate “new knowledge”, pointing to what he describes as published results resolving “almost 400” important open mathematics problems, with the wider maths community reviewing the work for insights. He argues progress in maths offers a template for other scientific domains, but says biology and drug discovery remain constrained by data availability and by downstream bottlenecks such as regulatory filings and clinical trial processes.
Brockman discusses OpenAI’s internal and product-direction bets on always-on agents, describing “dots” as a proactive assistant that monitors calendars, inboxes and work context to surface tasks without explicit prompting. He also predicts a shift away from fixed user interfaces, with AI generating bespoke interfaces when needed while aiming for “no UI” as the default interaction.
On cybersecurity, Brockman says the “physics” of the field are changing and frames the current period as a “defenders window” in which high-end AI can help organisations harden systems before offensive capability becomes broadly diffused. He says OpenAI has a “billion-dollar commitment” to help frontline defenders access models, and claims the company redeployed about 25% of its production engineering staff to security work, using frontier models to probe infrastructure and fix issues.
He closes by reiterating OpenAI’s stated mission to ensure AGI benefits humanity, arguing humans should remain “in the driver’s seat” even as model capabilities surpass those of individuals.
