No Priors

Greg Brockman on why new knowledge from AI is now impossible to deny

Greg Brockman· President and co-founder at OpenAI
·~39 min·English·No Priors
AI SafetyLLMAgentsReasoningAI CompanyPolicy
TL;DR

OpenAI president Greg Brockman argues AI has crossed from giving answers to generating genuinely new knowledge — pointing to roughly 400 resolved open math problems — and says the same shift is about to rewrite the physics of drug discovery and cybersecurity, with the whole project being to keep humans in the driver's seat.

01Core Claim

Impossible to Deny

AI has crossed the line from doing checkable arithmetic to resolving century-old open problems, and Brockman says the question of whether it can produce genuinely new knowledge is now settled — the answer is yes.

And there's always been this question of is AI actually able to come with new knowledge? And at this point, it's just impossible to deny, right? It is definitively yes, possible.

— Greg Brockman, No Priors
Key Insight
The milestone is not that AI answered a question but that domain experts call the resolutions surprising — the bar moves from “can it compute?” to “can it see what we missed?”, and the old era of incomprehensible machine proofs like the four-color theorem becomes the thing being left behind, not the model.
02Human Leverage

The Leverage Goes Way Up

The point of the math results is not a machine handing you an answer but a human equipped with these tools reaching results that were unthinkable before — the leverage on what one person can do goes way up.

It's really what can a human equipped with these tools achieve that was unthinkable before. And I think it's almost like that the leverage of humans to do whatever you want to do. It's going way way up

— Greg Brockman, No Priors
Key Insight
Framing the win as human leverage rather than machine autonomy is a positioning choice: it keeps the person as the agent and the model as the instrument, which is the same move that later lets OpenAI tell its safety story as keeping humans in the driver's seat.
03Science

Humans Are Bad at Biology

Humans are, in Brockman's view, badly suited to biology while a neural net has no reason to find biology harder than language, so first principles say drug discovery should fall the way math did — once the data is there.

I think we're going to find that humans are incredibly bad at biology, incredibly bad at drug discovery.

— Greg Brockman, No Priors
Key Insight
The first-principles argument has a load-bearing assumption hidden inside it — that the data exists. Brockman concedes biology is data-bottlenecked, so “it should work” really means “it should work wherever we can get or generate the data,” which is why he keeps returning to wet labs, simulation, and even regulatory paperwork rather than to the models themselves.
04Policy

Prove the Physics Changed

The slow parts of drug development are rules tuned for human-speed discovery, so once you can actually prove the physics of discovery has changed, Brockman expects regulation to reorient around what is now provable.

But if you can really prove that the physics of drug discovery have changed, then of course it should trickle through to every part of the system.

— Greg Brockman, No Priors
Key Insight
Brockman is making a regulatory argument dressed as a technical one: “prove the physics changed” puts the whole burden on evidence, but it also implies that once the evidence exists, the slow stages of drug approval are a choice society is making, not a law of nature — which is a pointed thing for a model provider to say to regulators.
05Product

The Best UI Is No UI

Dots — an always-on proactive agent you simply talk to — points at a future Brockman states plainly: the best interface is no interface, with the UI generated on the fly to fit whatever you need in the moment.

I think that the best UI is no UI and I think in AI that is where we're going

— Greg Brockman, No Priors
Key Insight
What Brockman is really describing is an interface generated per request instead of designed once — ask how to assemble a bike and get a throwaway app for exactly that — and he is candid about the hard problem it creates: when there is no fixed UI, how do you even surface to a user that some capability (say, a feature an enterprise admin has turned off) is missing?
06Cybersecurity

The Defender's Window

Brockman says the physics of cybersecurity is changing and we are in a defender's window — a period where frontier AI can harden systems before the same power diffuses to attackers — and defenders have to use it before it closes.

I do think the physics of cyber security are changing and we're in what we call the defenders window

— Greg Brockman, No Priors
Key Insight
Calling it a “window” quietly concedes that it closes — the same frontier capability that lets defenders get ahead is the capability that diffuses to attackers, so the edge is temporary and only real if defenders move now, and OpenAI's billion-dollar commitment is a bet on being the party that defines what moving now looks like.
07Security Economics

Security Should Outstrip Software

Because AI is driving the cost of writing software toward zero, Brockman argues the value moves to software you can trust, so security spending should end up vastly outstripping software spending — far above the low-single-digit share of IT budgets his interviewer floats as today's norm.

cyber spend, security spend should actually vastly outstrip software spend.

— Greg Brockman, No Priors
Key Insight
The claim only holds if software is nearly free — Brockman's premise is that AI collapses the cost of producing software to almost nothing, so spending shifts from making software to being able to trust it, which reframes security from an overhead line item into the thing you are actually paying for.
08Worldview

Humans in the Driver's Seat

Across science and security the constant is control: Brockman insists humans stay firmly in the driver's seat, keeping judgment and values even as models surpass them on most axes — more like a CEO directing expert teams than a worker being replaced.

We think that humans should be firmly in the driver's seat, should have control over the future, and this technology should amplify human agency.

— Greg Brockman, No Priors
Key Insight
The iMac-handle story is doing quiet work: it grants that the machine may be more capable on nearly every axis and relocates human worth to judgment and wanting rather than skill — “Are you happy? Is this what you wanted?” is the only question Brockman reserves for the human.