AI models like Google's Gemini improved accuracy on math problems from 20% to 80% after fine-tuning with chain-of-thought reasoning.

“there are some that are really much better at math than others uh there such as Google's Gemini recently did some fine tuning with what's called uh you know chain of of reasoning other words when you reason you go through a sequence of steps and when you solve a math problem you go through a sequence of par of steps of doing you know fitting first finding out what missing and then adding that and it went from 20% correct to 80 right on on those problems”