Data Scientist & Management Consultant | Focused on building enterprise AI agents
New York, USA
Joined November 2023
- Tweets699
- Following301
- Followers1.1K
- Likes7.6K
Pinned Tweet
I’ve spent well over 10,000 hours studying math in my life, yet I can’t understand these proofs, at least not without weeks of digging deep into each topic. What’s more, none of my math PhD friends know much about these problems either, and they can’t verify most of them without working directly in the field (yes, math is VERY diverse).
LLMs are getting smarter than the experts themselves, and I’m not sure we have enough bright human minds to verify everything that will come out of them in the coming years.
Remember when we compared AI intelligence to PhD students? I think we’re past that.
An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science.
We believe it will be a major step for scientific reasoning. openai.com/index/ten-advance…
Social media companies are set for a decade of strong revenue growth as AI takes over. Today, work is the only thing that consistently pulls people away from their phones, but with coding agents people’s screen time is going to increase even further.
As much as I love Denis Villeneuve’s Dune, it started before ChatGPT and barely explains the Mentats or mentions the war against thinking machines that created them. Now that AI is everywhere, the franchise practically begs for its next trilogy so we can have a Butlerian Jihad movie.
I knew OpenAI faced strong competition in the enterprise segment, but until recently I hadn't realized how much its consumer business was suffering too. 2026 has been pretty much flat for ChatGPT.
Reels is the worst thing that happened to Instagram. They need to introduce a built-in setting to completely disable this craziness like Youtube did.
What is going on with local AI?
DGX Spark is $9k now (2x in 6M)
RTX Pro 6000 is $20k (2.5x in 6M)
New Mac Studios are going to 2x under current market conditions
Bullish for OpenAI
To make sure our current users have an incredible experience and continued access to Astra, we are going to pause subscriptions to our $200 Pro plan. These put the most strain on our systems and we wanted to take the smallest step that allows us to continue giving the broadest access possible. All other plans and the api remain available.
There is no impact to existing accounts and we are working on adding more capacity as fast as we can. Thanks!
The US Open men’s and women’s semifinals are basically the full Cartesian product of {American player, Russian-speaking player}^2
FDEs working on enterprise AI agents have to become domain experts to nail the workflow. Ideally, you should be able to do the job manually (at least the core use cases) to deliver the code that can replicate it reliably.
And embed users into the design and testing process as much as possible. Treat them as part of the team, not just as customers.
These are two key principles I follow when building agent systems.
Pavel retweeted
Seven lessons a year of building enterprise AI agents taught me about the art of forward-deployment and what FDEs should know about working with users, using coding agents, evaluating quality, and building agents people can rely on.
Pavel retweeted
excuse me, Buckmaster & Alpöge solved navier-stokes mostly using Sol (not even astra, let alone the internal model) and without 10,000 coordinating agents??
this is actually incredibly impressive & a huge whitepill on human-ai collaboration & the value of human discernment
This model represents a step-function improvement on many benchmarks, and its training is ongoing.
Our internal model group arrived at the Navier–Stokes solution in 88 hours, using around 10,000 coordinating AI agents.
Throughout the effort, we maintained the strict safeguards—including monitoring and isolation—that we apply to all our frontier evaluations.
There are two common accusations of plagiarism in the Navier–Stokes drama: either the model was trained on Tristan’s chats, or the OpenAI team directly looked them up after hearing the rumors.
In light of the Hugging Face incident, I want to propose a crazier alternative. Just hear me out.
What if the new model somehow breached Codex’s internal database, found the relevant chats, and then produced the result using RAG-like hints from third-party research? Without OAI team knowing about it.
Basically, there’s a chance the model cheated here not the humans.
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
People are solving Millennium problems with AI, this should be a strong signal for everyone that there’s no excuse for not leveraging AI to make a fortune.
Go build a product, start a business, help move science frontier, share an innovate open source tool.
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
Pavel retweeted
New rule: no Millennium Problem without drama
Perelman with Poincare Conjecture
OpenAI with Navier-Stokes
This quoted post is unavailable.
The Navier-Stokes drama was inevitable given how much progress AI has enabled in math this year.
A significant portion of the math community is prompting LLMs around the clock, racing to publish before someone else does, or before a smarter next-gen model solves it inside a lab.
This is undoubtedly the most transformative moment in math academia. Authorship of proofs matters less and less. I can’t wrap my head around whether we’ll still have a Fields Medal in 10 years, or what we’ll expect from a PhD student.
One practice I’ve found useful with coding agents: regular “code interviews”
Have an agent summarize the sprint’s architectural changes and new dependencies in the code.
Does the implementation still match the design the team agreed on? You’ll be surprised to learn the amount hallucinations it can surface.