The Endless Exam: Mathematical Constructions from Today's Models toward Superintelligence
Endless Exam scores mathematical constructions without a cap; nine models fail to beat 30 published frontiers.
Endless Exam is a benchmark of fourteen parameterized families of mathematical construction problems, with automatic validity checks and relative quality scores that are not capped at 1. Challenges come from open problems, and larger instances are generated by varying parameters; compact certificates let large constructions be verified without listing every element. Nine models were evaluated on 69 distinct instances. Continuous scores separate performance, but none of 30 published-frontier references is surpassed. Generators, verifiers, references, model responses, and analysis are released.
- Fourteen parameterized families of mathematical construction problems.
- Quality scores are uncapped relative to published frontiers.
- Nine models on 69 instances; none beat 30 frontier references.
- Generators, verifiers, references, and model responses are released.
From the original article
1 min read at huggingface.co
We introduce the Endless Exam, a benchmark spanning fourteen parameterised families of mathematical construction problems, with verifiable scores that distinguish progress before and beyond published mathematical frontiers. Each submitted object is checked automatically for validity and assigned a relative quality score against a published frontier or construction baseline, without capping improvements at 1.
By Muhan Zhang · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.