LLM learn to solve non -complex mathematical problems

LLM, large language models, demonstrated excellent performance in the language, generation of text, code synthesis and many other tasks. However, mathematical reasoning is still too tough for them.

Recently, the developers of AI taught him to solve some mathematical puzzles, writes Xrust. On the victorious path there were thin settings using synthetic data. Experimental results with the Open-Llama-3B model on three different test sets show that the model gives out the correct solution.

II gave the following set of numbers: 34, 18, 31. 41, 19, 55. It was necessary to get «-110».

.

.

It should be noted that the lack of special mathematical training has become the main obstacle to the development of the solution algorithm. However, the problem was solved.

Xrust llm learn to solve non -complex mathematical problems