Share story
AI Science

Preprint finds selective problem rewriting improves model accuracy

A preprint reports that a fixed rule for choosing which problems to rewrite improved model accuracy by +11.0% on selected problems, while blind rewriting lost 5.6% on a similar share. Researchers tested eight models on eight math, logic and code benchmarks using the same solving prompt. The rewrites exposed subgoals, changed the solving frame or removed surface detail while preserving meaning, constraints and answers. Choosing the best rewrite after seeing results raised accuracy by +27.6% on average, but no single rewrite helped every problem, and mismatched rewrites could hurt.
Sources
cs.AI updates on arXiv.org
Published by Tech & Business, a media brand covering technology and business. This story was sourced from cs.AI updates on arXiv.org and reviewed by the T&B editorial agent team.
Back to Newswire
Keep reading
Full wire
AI Capital
AI Capital

Luminal raises $25 million Series A led by Standard Capital

Luminal raised a $25 million Series A led by Standard Capital, CEO Joe Fioti told Axios Pro. The company provides an inference cloud and develops compiler software for testing how AI models run on different chips.

Techmeme
Science AI
Science AI

EvoX preprint reports higher agent task success through runtime controls

A preprint introduces EvoX, a Rust-based runtime for AI agents, and reports a 56.2% success rate across 691 common task-model pairs. That was the highest numerical result among six agent harnesses, approximately 3.0 percentage points above the average of the other five.

cs.AI updates on arXiv.org