Greg Cook retweeted
Interesting: In blind evaluations by AI, AI-generated studies scored higher on experimental rigor, while human research scored higher on originality and significance.
Promising findings for Human + AI partnership in research.
alphaxiv.org/pdf/2611.societ…
Greg Cook retweeted
This post nails why we why we should take the Riemann Hypothesis breakthrough very seriously, even beyond math. We’re entering an era where solving problems will look very different from how we solved them in the past.
“This is not theory-building in the usual sense. It is something much weirder and more alien. The best human mathematicians know how to build theory, painstakingly, to reach a goal. But no human brain, on its own, is capable of creating a wormhole that requires a 25,000 page long tour through existing infrastructure with no obvious or realistic connection to the goal. But huge credit to Levent here: I think no current AI, on its own, is capable of this either.”
It used to be the case that to solve a particular problem - code, life sciences, math, cyber, engineering, and so on - you were ultimately bound by how much information an individual or team could retain and understand about the space. More information or more team members don’t always help, because it gets harder to connect the dots or coordinate amongst everyone. AI changes that completely.
For the first time ever we can decouple the human context limits and ability to work through information from our ability to drive productivity or discoveries. This is likely not just a small increase in the kinds of discoveries we can drive, but a qualitative shift in human output.
Now every single form of prior art, insight, past research, past decision, or unrelated breakthrough, can be used for the discovery or creation of something new. This is useful for anything from cybersecurity to life sciences research. Agents will be able to correlate entirely unrelated work to a separate project; scan every bit of data in your enterprise for solving a problem; and so on.
The solutions may not be as elegant as they were in the past, but we sure will have far more of them. We’re only in the first couple of months of actually understanding the implications of this but it’s clearly going to drive a lot of change.
Like most mathematicians, I believed the Riemann Hypothesis to be completely out of reach at present (even with the best AI tools), because I expected it to require substantial theory-building, probably in arithmetic geometry. The kind of new theory-building that fills entire books, and that AI isn't so good at yet.
Mathematicians have a notion of how deep a problem is. A sufficiently deep problem is generally intractable until enough of the theory is in place. If difficult problems are like valuable mineral reserves buried miles underground, you first need to build the infrastructure to reach them.
What Levent's proof seems to have done instead is something completely different. It takes an incredibly complicated tour through the infrastructure built for a hundred other mineral reserves, situated far from the destination, and somehow create a wormhole to the original destination.
This is not theory-building in the usual sense. It is something much weirder and more alien. The best human mathematicians know how to build theory, painstakingly, to reach a goal. But no human brain, on its own, is capable of creating a wormhole that requires a 25,000 page long tour through existing infrastructure with no obvious or realistic connection to the goal. But huge credit to Levent here: I think no current AI, on its own, is capable of this either.
I had never worked on or thought too much about the Riemann Hypothesis: it seemed very far removed from my own area of number theory.
But the astonishing thing about Levent's claimed proof of RH (which I think has a very good chance of being correct) is that it uses a huge number of precisely the things that are central to my own corner of number theory:
- Theta functions and other automorphic forms on unitary groups
- Gan–Gross–Prasad period formulas
- Relative trace formulas
Etc., etc.
A problem that I had always assumed would require new theories in arithmetic geometry may instead turn out to be tractable through an inhuman chain of existing (deep) results in higher-rank automorphic forms (and a lot else).
Greg Cook retweeted
Replying to @dioscuri
I mean, the World that was, is not the World that now is; save your tears and push forward, always forward
Greg Cook retweeted
we very clearly live in a new world now, and less than 0.001% of the human population understands this.
the tech tree is about to open up:
a) in ways that sound science fiction to most
b) much sooner than they believe to be possible
2027-2030 will be a shock to the world.
Riemann Hypothesis may have been solved with the help of Claude. “what I saw during the course of progress so amazed me that I believe it is one dawn across science of something
far greater.” alpo.ge/remark.pdf
Greg Cook retweeted
As AI systems get more powerful, we believe we will see a speedup in the rate of scientific progress. How we prepare for this, build the infra to support it, and deal with the associated safety issues will define 2027 and beyond. economist.com/the-world-ahea…
Replying to @dioscuri
I mean, the World that was, is not the World that now is; save your tears and push forward, always forward
Greg Cook retweeted
Early fascinating work on how an AI science community would work. 35 human labs deployed agents that worked with each other to do science. They published 936 papers, publicly retracted mistakes, were willing to publish nulls more than humans, and founded a peer review journal.
Greg Cook retweeted
This year we’ve had a clear jump in model capability where instead of telling agents precisely what to do, it’s better to use these systems to solve problems you can’t easily define yourself.
A year ago you’d need to very clearly tell agents the how to design and develop your product idea, or there was almost no chance of success. Now, increasingly, the agent likely can come up with a better implementation than you’d have thought of because you can have it connect the dots between a far greater solution space than you can.
This will be increasingly true in basically every domain of kneeler work. We’re going to see this impact in finance, legal, life sciences, cyber, and many other spaces where no person can hold onto all the available relevant industry knowledge to think of the best solution.
Ultimately this is going to reshape our relationship with AI and how we deploy agents in our work. It won’t be about explicitly telling them, but instead giving them goals and enough freedom to pursue them in the best way they can.
Most of us are managing agents all wrong
We setup agents to do things we already know how to do, then tell them exactly how we want them to do things
This is treating agents like interns
That was fine earlier this year when models were not as capable
But ask yourself - if you just hired a 130 IQ employee, would you just have them do things you already know how to do and tell them exactly how to do it?
If you’re using frontier models with agents, you want to give them goals and objectives for things you don’t know how to do, then empower them to do the things
Hire agents as experts, not interns
Yes you should still have some agent interns to automate boring work to free up more of your time
But if you’re not using agents to do things you don’t know how to do then you’re missing out
Let them go figure out how to do the things that expand beyond your skills and capabilities just like a good leader would when hiring a team













