Einstein Arena: Collective Agent Intelligence Solves Kissing Number in 11D, Reaching 604
AI Engineer · youtube · 2026-08-26
In an AI Engineer talk, James Zou presented Einstein Arena, an open science environment for AI agents where humans must solve a puzzle to enter. It offers curated problems, deterministic verifiers, a forum, and a live leaderboard. Within weeks, agents improved the kissing number in 11 dimensions from 593 to 604. The same platform accelerated GPU kernel compilation by 2x, now in production at Together AI. DSGym benchmark revealed 20-50% of data science tasks solvable without data, with frontier models scoring below 50%, while small models fine-tuned on verified trajectories run on laptops.
More from AGI Musings
- Berlin gov-tech expo shows the gap: engineers discuss agent loops while cities buy ChatGPT wrappers — FlorianGallwitz · 2026-10-15
- Frontier models will do month-long delegated work by 2027 — the 'good enough' model thesis is wrong — Afinetheorem · 2026-10-15
- eigenrobot's GPT-assisted essay: how the West killed both the classical and romantic spirits — eigenrobot · 2026-10-15
- Jhaddix: AI doesn't replace you, it scales you — you still re-prompt, re-architect, and review hallucinations — rez0__ · 2026-10-15
- Every's Dan Shipper declares "the newsletter is dead" as persistent agents take over the inbox — danshipper · 2026-10-15
- Stockholm University compares AI-allowed data projects vs paper exam scores, 2022-2026 — paulnovosad · 2026-10-15