Fetching from the wire…
01
02
03
04
05
06
07
08
Vibe Coding2026-09-19 · source-backed
The conjecture on omnific integers, open around 50 years, with verification recorded in the Palomar registry. Early attempts produced grandiose but incoherent mathematics. What worked: separating Lean formalization of peer-reviewed sources from exploration of novel results, forcing standalone Lean files importing only Mathlib so output stayed auditable, running specialized agents (PM, math researcher, red team, formalizer), and twice discarding all accumulated work to refocus. overreacted.io The transferable part is the auditing structure, and "twice threw away everything" is the honest detail most writeups omit.
Each link below shares sources, entities, or timing with this story.
Two facts sit next to each other and neither cancels the other out. Anthropic published on September 4 that an internal general-purpose research model, roughly comparable to Claude Fable 5.1, formalized Fermat's Last Theorem in Lean over 11 days working largely autonomously. T...
openai/PrimeGaps186, created September 2 under Apache-2.0, at 117 stars, formalizes lim inf(p_{n+1} - p_n) ≤ 186 via the Dickman-Hardy-Littlewood conjecture applied to a 40-element admissible tuple. It proves three theorems but remains conditional on a Kloosterman3 bound from...
His August 2 post concedes the result is real and attacks the inference as a fallacy of composition: success on one form of fancy cognition doesn't mean success on all forms is imminent. His sharpest technical objection is that math is uniquely favorable because it "allows for...
Victor Taelin's Bend 2 went public on September 17 and took 502 points on Hacker News. The repo is at 21,032 stars. The mechanism: you write theorem statements in a LAWS.bend file, for example law add_zero: for x: Nat {Nat.add(x, 0n) == x : Nat}. The AI writes the discharging...
Terry Tao hosted a September 12 guest post by Silvia De Toffoli and Eamon Duede attacking two assumptions behind the coverage: that AI solved a mathematical problem, and that mathematics is only about solving problems. Their split is between logical validity and intelligible u...
Created September 6, its stated split is "The LLM interprets intent; the MathKernel establishes mathematical evidence" (GitHub). It orchestrates SymPy, mpmath for arbitrary-precision and interval arithmetic, Z3, Lean, SciPy, NumPy/CuPy and Numba behind one typed MathIR, speaki...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.