arxiv Moonshot AI (arXiv) · 550d ago

GAIR-NLP releases LIMO, achieving strong math reasoning with 817 samples

from English

GAIR-NLP introduces LIMO, a model demonstrating that complex mathematical reasoning can be effectively elicited using only 817 curated training samples. This approach challenges the conventional assumption that supervised fine-tuning requires massive datasets to avoid memorization.

LIMO achieves 57.1% accuracy on the AIME benchmark and 94.8% on MATH.
The model uses just 1% of the training data required by previous strong SFT-based models.
It demonstrates exceptional out-of-distribution generalization, improving performance by 40.5% absolute across 10 diverse benchmarks.
Results outperform models trained on 100x more data, supporting the "Less-Is-More Reasoning Hypothesis."

The authors propose that sophisticated reasoning capabilities emerge when foundation models leverage rich pre-trained knowledge through minimal, precisely orchestrated cognitive templates.

Importance 2/3 Moonshot AI (arXiv) Zhipu AI Reasoning models Research paper

Benchmarks

Benchmark	Model	Score
AIME 2024	LIMO	57.1%
MATH-500	LIMO	94.8%

Read original

arxiv arXiv cs.CL · 30d ago

Probing Eternis-Forecaster 8B reveals internal representations improve calibration and expose CoT faithfulness issues

Researchers investigated whether internal representations in large language models provide a more direct window into calibration and faithfulness than chain-of-thought (CoT) reasoning. Working with Eternis-Forecaster 8B on OpenForesight, they trained representation-pooling probes on intermediate activations that achieved substantially better calibration compared to the model's standard outputs.

media Don't Worry About the Vase · 5d ago · 5 views

OpenAI's Astra solves 10 major open mathematics problems

OpenAI has released results from its unreleased internal model, Astra, which successfully solved ten significant open problems in mathematics. The solutions were generated by the model and subsequently formalized into Lean certificates by human researchers.

blog Simon Willison · 7d ago · 12 views

OpenAI's Astra model solves ten long-standing math problems using GPT-5.6 Sol

OpenAI has released solutions to ten mathematical and theoretical computer science problems that had seen no progress on their main results for at least a decade, using an internal version of its next major model, Astra.

arxiv arXiv cs.CL · 12d ago · 2 views

PIVOT shares prefix scans across query groups to accelerate token-level sparse attention

PIVOT (Proxy Indexing Via One full-prefix Traversal) is a training-free, drop-in replacement for the DeepSeek Sparse Attention (DSA) indexer that reduces computational redundancy by sharing one prefix scan across a group of nearby queries. Instead of scoring every preceding token for each query individually, PIVOT aggregates the group into a single proxy query to obtain a candidate set, from which top-k tokens are selected for each query.

arxiv arXiv cs.CL · 12d ago

Zing framework improves LLM social intelligence via SoMBench benchmark and Actio grounding

The report introduces Zing, an integrated framework designed to enhance the social intelligence of large language models by measuring, internalizing, and grounding social capabilities. The authors present SoMBench, a psychology-grounded benchmark spanning 17 secondary dimensions and 3,481 expert-verified instances, which reveals that current LLMs have significant room for improvement with no model reaching near-ceiling performance.