#rstarmath — Public Fediverse posts

Microsoft's rStar-Math aims to enhance small language models' math reasoning by:
➡️ using System 2-style deep thinking (careful analysis and step-by-step reasoning).
➡️ Monte Carlo Tree Search and self-evolution to generate high-quality training data.

It appears to achieve 90% accuracy on the MATH dataset and gets close to larger models, making advanced AI more accessible.

https://www.marktechpost.com/2025/01/10/microsoft-ai-introduces-rstar-math-a-self-evolved-system-2-deep-thinking-approach-that-significantly-boosts-the-math-reasoning-capabilities-of-small-llms/

#AI #MachineLearning #Math #DeepLearning #ArtificialIntelligence #DataScience #rStarMath

@emollick.bsky.social

Paper shows very small #LLMs can match or beat larger ones through 'deep thinking'-evaluating different solution paths. Their 7B model beats o1-preview on complex math by exploring 64 different solutions & picking the best one.
Test-time compute paradigm seems really fruitful.

Source : https://x.com/iruletheworldmo/status/1877993301896294482?s=46

#o1 #o3 #rstarmath #llm #openai #msft

Microsoft's new rStar-Math framework sets new standards in mathematical reasoning benchmarks using Small Language Models #AI #Math #SLMs #AIResearch #Microsoft #rStarMath #AIBenchmarks #AIFrameworks #COT #CHainOfThought

https://winbuzzer.com/2025/01/10/microsofts-rstar-math-framework-lets-small-ai-models-outperform-openais-o1-series-xcxwbn/