Post-Cutoff.com
  1. Home
  2. Timeline
  3. 2026
  4. EPFL's LinCodeEvolve (Viazovska, Abbe) finds seven…

EPFL's LinCodeEvolve (Viazovska, Abbe) finds seven record-breaking binary linear codes with LLM-guided program search

★★★after cutoffscienceEPFLconfidence: high

On Sept 29, 2026 EPFL researchers Amal Seddas, Vladyslav Shashkov, Maryna Viazovska (Fields Medal 2022) and Emmanuel Abbe posted LinCodeEvolve, an LLM-driven evolutionary program search (built on ShinkaEvolve and EvoTune) that found seven binary linear codes beating the best-known minimum distances in Grassl's CodeTables, e.g. [172,21,66] and [200,21,77]. With standard modifications they improve 22 table entries. Every code was verified by exhaustive enumeration, and the table maintainer checked them. The work applies the FunSearch/AlphaEvolve approach to a classic coding-theory benchmark.

Key facts

Science result

Field
mathematics / coding theory
Problem
Improving best-known lower bounds on the minimum distance of binary linear codes (Grassl's CodeTables)
Result
Seven new record binary linear codes (lengths 172–200, dimensions 20–22), improving 22 table entries after standard modifications
AI system
LinCodeEvolve, GPT-6 Astra
Human role
Human-designed LLM-guided search with expert supervision of strategies; codes verified by exhaustive enumeration
Verification
Exhaustive computer verification; checked by the CodeTables maintainer
Status
confirmed

What happened

Finding binary linear codes with large minimum distance is a central coding-theory problem, and certifying minimum distance is NP-hard. LinCodeEvolve keeps ShinkaEvolve's island-based archive, novelty judge and model selection. Candidates enter the archive only with an exact certificate (full weight distribution). Among codes with equal distance it prefers fewer minimum-weight codewords.

Why it matters

Viazovska, who solved sphere packing in dimensions 8 and 24, is now co-authoring LLM-search papers. It is another case of FunSearch-style search improving a long-maintained table of records with modest compute (one H100). The authors note that careful prompting of a frontier model (GPT-6 Astra) came close on some parameters, but systematic search did better across many parameter pairs.

Changelog

  • 2026-09-30: created (sweep 2026-09-30, arXiv AI-disclosure section)

Related events

  1. AlphaEvolve: Gemini-powered agent discovers new algorithms ★★★★
  2. Summer 2026 flood: dozens of named conjectures settled on arXiv with disclosed AI help (July–September catalogue) ★★★★

Sources (2)

id: 2026-09-29-lincodeevolve-record-binary-codes · updated 2026-09-30 · open in the interactive timeline