
We are poised to translate all known math into formal code.
✓szegő assistant professor of mathematics @Stanford

We are poised to translate all known math into formal code.

"AI struggles to understand its own mathematical reasoning, but becomes much better at building on its ideas when humans simplify them."


Nice distinction: > arXiv for distilled/digested papers > Hexagon for raw materials

I am seeing a lot of confusion here. The main point is that in the next years, during the AI takeoff, digital life will take up more surface area of the modern world. Then *within the context* of whatever digital is instantiated in code, it can be inspected and verified by swarms of agents in a formal language. And to reiterate @AlexKontorovich's qualification "There can still be arguments / lawsuits about whether variables were properly assigned, etc. But after accepting the assignments, the rest is deterministic." That is to say, this decreases the surface area of: what is subject to debate vs. what is matter-of-record, to which an impartial protocol is applied.


Pacing the frontier

If law can be formalized, this will be truly transformative for socitety.

🚨HOLY SHIT. Just weeks after OpenAI announced new AI math discoveries, Google has dropped an INSANE list of its own mathematical breakthroughs using Gemini 4 Argon, Carbon, and Hydrogen agents. And one of them dates back 99 YEARS… >Birkhoff conjecture: open since 1927 >FIRST machine-checked formalization And that's just ONE of the breakthroughs… >3D Bochner–Riesz conjecture: 50+ YEARS first formalization >AI finds a GAP in the informal proof then FIXES it >C¹ regularity problem: open for ~20 YEARS >Google reports RESOLVING it >Jamison Caterpillar Conjecture: 40 YEARS AI verifies UCLA researchers' proofs But it gets even crazier… >Weil conjectures: last proved in 1974 >Google's /teamwork-LEAP AUTONOMOUSLY formalizes them >1.8 MILLION lines of machine-checked Lean >28,800 subagent invocations And there's MORE… >Courtade–Kumar conjecture: proposed in 2013 >Gemini finds counterexamples to a broader conjecture >humans reduce the problem to 14 variables >AI develops most of the new mathematical bounds >and verification certificates 50 MILLION lines of formally checked Lean 4 proof code All through Google's Gemini-powered multi-agent research systems, working alongside human mathematicians. This is INSANE.


Very nice video with @AlexKontorovich is the @BrianGreene the quasi-Riemann Hypothesis. Alex is a great expositor for the public: you might say he's the Brian Greene of math youtube.com/watch?v=ljFZfYZL…

In 1966, Paul Cohen won the Fields medal for showing the independence of: the continuum hypothesis and the axiom of choice. Gödel reacted to Cohen’s Fields medal by saying that he had already proved the result in the 1940s. Yet when pushed for a proof by his peers, Gödel was unable to reconstruct one from his notes. "The first task von Plato set Claude was to discover if Gödel’s workbooks had a proof of the independence of the axiom of choice, which would reveal that Gödel had been telling the truth in 1966. The answer came back: yes, almost. “He was just one step away,”"

"Within a couple of days, Claude had provided a comprehensive digest of the entire material, revealing Gödel had proved hundreds of theorems that no one knew about."

The true schrodinger's cat is the status of yang-mills

Claude is multilingual: The Poincare conjecture formalized, first in Lean, now in HOL Light
