Single Digit Tokenization improves LLM Math abilities by up to 70x

Re:Single Digit Tokenization improves LLM Math abilities by ... by vbezhenar on Thursday October 19, 2023 @01:12AM (#37938746)

I don't understand how this is useful. People don't calculate numbers in brain, they learned to use calculator. LLM should just learn to access external API every time they need to make a calculation instead of spending precious neurones on primitive tech already solved by transistors.

[ Parent | on HN ]

Re:Single Digit Tokenization improves LLM Math abilities by ... by alexcannan on Thursday October 19, 2023 @01:33AM (#37938870)

I'm personally excited about LLMs becoming more number-fluent for regression tasks, e.g. read this paper and give it a grade 0-100

[ Parent | on HN ]

Re:Single Digit Tokenization improves LLM Math abilities by ... by vessenes on Thursday October 19, 2023 @12:55PM (#37945365)

Well the predicting tests they do would not be possible in this scenario, because they want to finetune a model against a set of data in order to make those predictions, and API calls aren't differentiable -> they can't train -> they can't use it to learn to predict. So, I think they see a use case here.

[ Parent | on HN ]