Author(s): Dulan Jayawickrama

Originally published on Towards AI.

I Don’t Let LLMs Do Even Simple Math (And You Shouldn’t Either)

Ever wondered why your trusty AI buddy flubs even the simplest math? I sure did. Last week I discovered a bug in an AI service I was working on: users could ask, “What’s my total usage in the past month?” Then we have to add 30 values, but our AI kept miscalculating the sum.

The article discusses the unreliability of large language models (LLMs) in performing even simple arithmetic calculations, as demonstrated through the author’s experiences with a faulty AI service. It offers two solutions to bypass the limitations of LLMs: first, by having them write Python scripts for calculations which users can execute, and second, by utilizing a calculator tool that the LLM can call to perform arithmetic tasks safely. The author emphasizes the importance of safety and accuracy, advising users not to place their trust in LLMs for numerical computation.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI