Robot Ventures

Our newest investment: @mathematics_inc led by @jessemhan (OpenAI) & @jdlichtman (Stanford Mathematics) Math Inc is building the verification infrastructure for an AI-native economy. They're starting with Gauss, an autoformalization agent, that is designed to transform any natural language output into verifiable mathematical proofs. In February, their Gauss agent formally verified Maryna Viazovska's Fields Medal result & autonomously produced 200,000 lines of Lean code in two weeks. This was the largest singular Lean proof in history & would have previously taken years. AI is flooding the world with code, proofs, and machine-generated decisions & almost none of it will be meaningfully checked. Agents deploy faster than humans can audit. Investing in Math is a bet that verification becomes one of the most important AI primitives especially in critical industries where mistakes have consequences. My friend @ani_pai wrote a great piece on why they're investing in @mathematics_inc too:

@0xmadisonn walked so the agents could “sprint”

They said the quiet part out loud 🤖

Enjoyed this. Everyone is focused on making agents more intelligent. But intelligence doesn’t move money: Acceptance does. The moment work becomes payable is the moment value becomes real. And whatever system controls that moment controls the economy imo. This won’t look like a marketplace, but will look like an underwriter. Credibility becomes the primitive.

Dope article! I think what this leads to is an 'economic harness' for agents that contains history/ reputation relevant dat and various forms of constraints to provide custom economic privileges such as delegation and undercollateralized credit.

Math, Inc. is proud to announce an all-star group of Veritas Fellows: Renowned professor Kevin Buzzard, alongside Fields Medalists Maryna Viazovska and Terence Tao. They will lead teams to build formal mathematics at unprecedented scale. 🧵

You are early.bulk.trade.

"It is a truth universally acknowledged, that an AI bot in possession of good credit must be in want of a wife" - Jane Austen

Credit Rules Everything Around Me

Great post. The human-agent relationship is going to get rebuilt dramatically over the next year. Several core primitives (as outlined in this post) are still needed to create the future of work. Startups take note. ✍️ x.com/0xmadisonn/sta…

Great read.

the two most overrated ideas in agentland: proof of human and agentic marketplaces one is identity theater, the other is a demo w/ a biz model stapled to it @tarunchitra and i wrote the equation for what actually prints robvc.com/research/machi…

Using Gauss, this is the first FrontierMath open problem to not only be solved but also autoformalized within hours of the solution being discovered. github.com/math-inc/Front… The formalization is native_decide-free and validated with @leanprover’s comparator. In addition to the uniform lower bound H(n) ≥ 26/25 * k_n for n ≥ 15, Gauss also verified the asymptotic lower bound lim inf H(n)/k_n ≥ 2ln(2). Combining with the upper bound in prior work of Brian and Larson, this yields H(n) ~ nln(n) as n → ∞. This is essentially the optimal result for this problem, and autoformalization greatly reduced the time required to validate the solution. In particular, Gauss was able to establish all the necessary theory within a few hours of the announcement of the initial informal attempt.

my baby got a facelift and now she doesn’t recognize her old friends 💅🏽

"As AI systems generate millions of lines of code, mathematical proofs, and scientific designs, the bottleneck around scale becomes trust." #LeanLang #LeanProver #FormalVerification

"I started this journey with the dream that one day human mathematicians would work alongside artificially intelligent ones. I saw a future where machines would search for proofs at a planetary scale, eventually going beyond proof search to synthesize compelling definitions and abstractions in a self-reinforcing loop, and where one day mathematics would become a shared, formally verified, and mostly autonomous computational construct, like the operating system of a mainframe computer whose details are hidden from the end users." These were the first words of my PhD thesis, from nearly four years ago. My PhD advisor, Tom Hales, revolutionized the field of formal mathematics with the Flyspeck project. The ideas in this manifesto speak to one of the deep motivations for early herculean formalization projects like Tom's - after discovering methods that were so alien, new, and written at inhuman scale that human reviewers and the institutions sponsoring them couldn't keep up, Tom spent over a decade blazing a trail to determine the truth for himself and point the way for the rest of the field. @mathematics_inc is now ushering in an age where efforts like that will take days instead of decades and where anyone who dares to have courageous ideas will be able to see whether those ideas are true. We are thrilled to partner with @tarunchitra @ani_pai and @robotventures to embark on the rest of this journey to scale autoformalization to verified superintelligence and beyond.

Humanity advances when individuals gain the capacity to verify truth for themselves, without institutional permission. We're thrilled to partner with @robotventures on our mission to scale autoformalization, create verified superintelligence, and set the truth free.

Excited to share that we’re doubling down on @mathematics_inc. AI is generating more code than ever, but correctness isn’t guaranteed and failures can be catastrophic. Their agent Gauss delivers a missing primitive for AI: provably correct code and knowledge. Thrilled to keep working with @jessemhan, @jdlichtman, and the @mathematics_inc team!

Weeks ago: everyone got excited about the autoformalization of Erdös problems Now: Graduated a few orders of magnitude to the autoformalization of multiple Fields Medal results with generated code that is ~10% (!) of the size of mathlib RH will be solved in my lifetime 😀