BEIJING – A joint analysis group has developed a general artificial intelligence (AGI) system able to each autonomous downside proposing and automatic downside fixing, marking a vital milestone in the self-developed logic cores for automated reasoning.
In efficiency and purposeful variety, the system, TongGeometry, has totally outperformed worldwide benchmarks, together with DeepMind’s AlphaGeometry. This represents a serious step ahead in AI-assisted mathematical analysis and the localization of clever training.
The examine, collectively performed by the Beijing Institute for General Artificial Intelligence (BIGAI), the School of Psychological and Cognitive Sciences at Peking University, the School of Intelligence Science and Technology at Peking University, the Institute for Artificial Intelligence at Peking University, and the Wuhan Institute for Artificial Intelligence at Peking University, was revealed in the journal Nature Machine Intelligence on Monday.
READ MORE: Pioneering university cultivating top-tier talent in AI
Mathematics Olympiads have lengthy served because the litmus take a look at for AI’s logical reasoning capabilities. In early 2024, DeepMind’s AlphaGeometry made world headlines by showcasing AI’s monumental potential for problem-solving. However, AlphaGeometry is actually a “passive solver” whose coaching depends closely on large-scale artificial datasets and expensive computational assets.
In distinction, the independently developed TongGeometry displays a better dimension of intelligence. It isn’t merely an “honor student” able to scoring full marks, but additionally a “master teacher” able to creating elegant and novel mathematical issues.
“We identified a profound duality in our research: when the proof difficulty of a geometric proposition is far higher than its construction complexity, it possesses ‘aesthetic value’ as an Olympiad-level problem,” mentioned Zhang Chi, the primary creator of the paper and a researcher at BIGAI.
“By modeling this duality, TongGeometry can precisely capture high-quality problems that meet the aesthetic standards of human mathematicians from a vast pool of spatial combinations. This is a global first, representing a paradigm shift from ‘imitative solving’ to ‘autonomous creation’,” Zhang famous.
TongGeometry clearly highlights the prevalence of unique home know-how in phrases of efficiency. While AlphaGeometry requires huge computing clusters, TongGeometry can remedy all International Mathematical Olympiad geometry issues from 2000 onward in 38 minutes or much less utilizing only a single consumer-grade GPU.
Its reasoning effectivity and accuracy have reached world-leading ranges. Furthermore, the system makes use of an progressive normalized illustration know-how to compress the search house by a number of orders of magnitude, successfully fixing the trail explosion downside inherent in conventional strategies.
“The significance of TongGeometry lies not only in the increase in solving speed but in its realization of the ‘small data, big task’ paradigm by simulating the intuition and aesthetics of human mathematicians,” mentioned Zhu Yixin, assistant professor from the School of Psychological and Cognitive Sciences at Peking University.
“This path, which does not depend on massive labeled data but evolves through internal logic, is the key to the development of AGI. Our system not only benchmarks against the most advanced international AI but also leads the way in understanding the underlying aesthetics of logic and the autonomous discovery of scientific laws,” Zhu mentioned.
READ MORE: AI-driven large models showcase great application potential
Three new geometry issues autonomously generated by the system had been formally chosen for the 2024 Chinese Mathematical Olympiad (Beijing District).
This breakthrough gives core technical help for future advances in automated mathematical proofs, customized clever training, and the event of “Science Large Language Models.”
Going ahead, the joint analysis group will proceed to iterate on the “Tong” sequence of general intelligence fashions, driving Chinese AI know-how to take the lead in extra fields of advanced logic and scientific discovery.