CRAM hits ~100% Sudoku-Extreme accuracy with just 30 training puzzles vs 700+ for looped models
IgorCarron · x · 2026-11-17
Ben Hoover reports that CRAM needs far less training data than looped models: 100% test accuracy on Sudoku-Extreme with only 30 puzzles, versus 700+ for larger looped models. The key insight: learning constraints is much easier than learning reasoning trajectories. Igor Carron frames it as a new kind of phase transition, recalling past work searching for phase transitions in Sudoku-solving algorithms.
Related event: IBM's CRAM Energy Model Trains 1,000x Faster on Sudoku(3 posts)→
More from Research
- Jeremy Howard reimplements MIT's Recursive Language Models paper inside SolveIt, more concise than original — jeremyphoward · 2026-11-17
- CRAM claims 100,000x+ speedup on Sudoku — 'I read the paper, it's Sudoku' — willdepue · 2026-11-17
- SperidLabs' Iris-Z: One Unified Generative Model Beats Specialist Depth Models at Scale — speridlabs · 2026-11-17
- Vibe Checking the IRS: Auto-Formalizing the Tax Code in Lean to Catch Errors — SnooPeripherals5313 · 2026-11-17
- MIT distills Monet, Van Gogh and 14 other artists' entire oeuvres into single paintings at ECCV 2026 — alexbilz · 2026-11-17
- Topos Bio teams with NVIDIA cuPotential to boost molecular dynamics throughput up to 3.6x — AllThingsApx · 2026-11-17