Greg Brockman：主动代理、可理解的科学发现与安全防御窗口

Brockman presents the reported mathematics results as evidence that AI can generate new knowledge, while explicitly distinguishing solved problems from entirely new grand theories.

The conversation treats human-understandable insights and adversarial checking as complements to strong answers, rather than assuming that a fluent explanation guarantees a correct result.

Brockman argues that transferring AI progress from mathematics to biology depends on obtaining suitable data, including cases where new experiments or simulations are still necessary.

The drug-development discussion spans discovery, documentation, and later-stage bottlenecks, but Brockman says that replacing clinical trials with neural-network predictions is not a point already reached.

Brockman describes Dots as a change in form factor that combines an always-working agent with calendar, inbox, and project context to notice tasks without waiting for a fresh prompt.

Brockman expects consumer and workplace agents to converge, while emphasizing that compute economics and organizational controls create real differences in what users can access and do.

His no-fixed-interface vision still includes generated visual tools when useful, and it leaves an unresolved need to show users which capabilities are unavailable because of administrative restrictions.

Brockman's defender's window is the period in which leading AI capabilities can help authorized defenders fix systems before comparable capabilities become more widely available.

He describes repeated model-assisted security reviews and argues that reliability may attract more spending than cheap software creation, but he presents that spending relationship as a personal judgment.

Brockman's position on human agency keeps people responsible for goals and judgments even when AI specialists know more, so stronger capability is not treated as automatic authority over human preferences.