What is TeleAI-t1-preview
TeleAI-t1-preview is a "complex reasoning large model" released by the China Telecom Artificial Intelligence Research Institute, featuring powerful logical reasoning and mathematical derivation capabilities. Through reinforcement learning training methods, it introduces thinking paradigms such as exploration and reflection, improving the accuracy of solving complex problems. In the 2024 American Invitational Mathematics Examination (AIME) and MATH500 evaluations, the model scored 60 and 93.8 points respectively, surpassing benchmark models such as OpenAI's o1-preview and GPT-4o. It can accurately process classical Chinese problems from "The Nine Chapters on the Mathematical Art," converting them into modern Chinese and providing detailed derivations. TeleAI-t1-preview will soon be launched on the Tianyi AI Open Platform and will play an important role in fields such as education and scientific research in the future.
Main Features of TeleAI-t1-preview
- Mathematical and Logical Reasoning Capabilities
- Solving Highly Difficult Mathematical Problems: TeleAI-t1-preview performs excellently in mathematical reasoning and can handle complex mathematical problems. In the 2024 American Invitational Mathematics Examination (AIME) and MATH500 evaluations, it achieved high scores of 60 and 93.8 points respectively, greatly surpassing benchmark models such as OpenAI's o1-preview and GPT-4o.
- Graduate-Level Question-Answering Test: In the graduate-level question-answering test GPQA Diamond, TeleAI-t1-preview scored higher than GPT-4o, matching the performance level of Claude 3.5 Sonnet.
- Understanding Classical Mathematical Texts: It can process classical mathematical texts such as "The Nine Chapters on the Mathematical Art," first understanding and simplifying the classical Chinese problems, converting them into modern Chinese, and then performing mathematical derivation and solving.
- Thinking and Reasoning Capabilities
- Combining Figurative and Abstract Thinking: TeleAI-t1-preview can combine figurative thinking with abstract thinking, concretizing the scenarios involved in complex problems to assist in understanding the questions.
- Complex Strategic Reasoning: When facing extremely "brain-burning" strategic reasoning problems, it can quickly understand the game rules and solve the problem, listing its understanding of the game rules, analysis of scenario props, analysis of advantages and disadvantages, and providing a solution strategy.
- Conversion Between Ancient and Modern Units: When handling conversions between ancient and modern units, TeleAI-t1-preview can demonstrate rigor and ensure the accuracy of the answer.
Technical Principles of TeleAI-t1-preview
- Reinforcement Learning and Thinking Paradigms: The model adopts reinforcement learning training methods and introduces thinking paradigms such as exploration and reflection. It can optimize its reasoning ability through trial and error, greatly improving accuracy on complex problems such as mathematical derivation and logical reasoning.
- Data Preparation: The research institute collected and constructed a high-quality reasoning dataset centered on mathematics and supplemented by multiple disciplines, ensuring that the model can adapt to different types of reasoning tasks.
- Judge Model (Evaluation Model): A dedicated evaluation model was trained to analyze and assess the correctness of the model's long reasoning chains, providing precise guidance for the model's reflection and error correction.
- Supervised Fine-Tuning (SFT) Stage: Based on Monte Carlo Tree Search (MCTS), high-quality long reasoning data is constructed, combining the accuracy of each step and the solution length to select the optimal path. Through the Judge Model, low-accuracy paths are analyzed and corrected to construct high-quality chain-of-thought data for SFT training.
- Reinforcement Learning Stage: Additionally constructed a Rule-based Reward Model, providing sufficiently accurate reward signals to further enhance the model's logical reasoning ability through online reinforcement learning algorithms.
Application scenarios of TeleAI-t1-preview
- Mathematics learning and competition tutoring: TeleAI-t1-preview can handle complex mathematical problems, including high school mathematics competitions and graduate-level mathematics problems.
- Analysis of ancient mathematics problems: The model can understand and simplify ancient classical Chinese mathematics problems such as those in The Nine Chapters on the Mathematical Art, convert them into modern Chinese, and perform mathematical derivations, providing strong support for learning ancient mathematics.
- Logical reasoning and strategy analysis: TeleAI-t1-preview performs excellently in handling complex strategic reasoning problems, can quickly understand the rules and solve the problems, listing its understanding of the game rules, analysis of scene props, analysis of advantages and disadvantages, and providing problem-solving strategies.
- Interdisciplinary research support: Powerful logical reasoning ability can assist researchers, help solve complex logical problems, and improve research efficiency.