Focused on LLMs Subject Solution Method in Mathematics Education

The large language models (LLM) have been an open question for a long time to choose what type of training data while giving mathematical capabilities. A new research reveals that the traditional approach is significantly more effective than the**-based selection.

The researchers reached two main findings:

-Yöntem creates a difference in its similarity: Problems in different mathematical areas such as htimal and kombinatoric, if it contains the same solution steps, the model provides more effective transfer than examples in learning the target problem.

• Deneysel design: It makes it possible to verify the results using a balanced 2×2 design that compares work, possibilities and cobinatoric issues with various solution approaches.

• Fine-tuning results: After model fine adjustment (fine-tuning), the logic of the work made has been determined to provide higher success rate than the subject-based choice of benzene educational data.

This finding shows that language models are soyut logical structures consolidated from mathematical issues can be transferred independently. In practical sense, LLM developers can create more powerful mathematical models by selecting similar methods from a wide range of subjects while preparing a training set.

As a result, this research has the potential to change the strategy of developing mathematical intelligence with artificial intelligence models.