saved
The Changing Physics of Cybersecurity and Drug Development with OpenAI President Greg Brockman
Hraness wrote this summary from a saved copy of the source. Quotations are taken word for word from the source.
gist
OpenAI president Greg Brockman tells No Priors hosts Sarah Guo and Elad Gil that AI now produces genuinely new knowledge. He cites nearly 400 resolved open math problems and argues the same methods can crack biology where data exists. He describes Dots, OpenAI's always-on proactive agent, says the best UI is no UI, and argues defenders are in a time-limited window to use frontier models to find and fix vulnerabilities.
ideas
- Math is the proof point. OpenAI published almost 400 important open problems now resolved, and Brockman says mathematicians are finding surprising insights in the proofs rather than opaque case checks.
- Biology is data bottlenecked. He ranks drug discovery as the top application but says progress depends on data, wet labs and simulation, and AI can shrink slow steps like year-long regulatory filings.
- Agents become proactive. Dots runs continuously on its own cloud computer, tracks commitments across inbox, Slack and calendar, and flags conflicts unprompted.
- Defenders have a window. OpenAI put about 25% of production engineers on security until its peak model saturated the issues it could find, runs a Daybreak defender program, and committed a billion dollars to frontline defenders.
- Security should outspend software. As software gets cheap, Brockman argues reliable, secure software becomes the real value, so security spend should vastly exceed software spend.
quotes
“I think that the best UI is no UI.”
“I think we're going to find that humans are incredibly bad at biology, incredibly bad at drug discovery.”
“cybersecurity spend should actually vastly outstrip software spend”