program-aided language model

A prompting approach in which a language modellanguage model, n.: A system that has learned how likely different sequences of words are, so it can predict the next word in a text and, by doing that repeatedly, write text of its own. translates a reasoning problem into an executable program — typically Python that sets up variables and computes the answer — and an interpreter runs the program to produce the final result. The split plays to each side's strengths: the model reads the problem and works out the steps, the computer does the exact arithmetic, and the slips that creep into purely textual chain of thoughtchain of thought, n.: Getting an AI to write out its reasoning step by step before giving an answer, the way a student shows their working. This often makes answers to math and logic problems more accurate. are avoided. The idea extends to symbolic solvers and calculators, and it underlies the code-interpreter and code executioncode execution, n.: Letting an AI write a small program and actually run it — to do math, analyze a spreadsheet or test its own code — then use the real output instead of guessing. tools that models now call routinely while reasoning; reasoning modelsreasoning model, n.: An AI model trained to work through a problem step by step in hidden 'thinking' text before it answers. Letting it think longer generally makes it better at math, coding and planning. trained with tools learn for themselves when to write such code.

Switching to program-aided prompting fixed the rounding errors, because the interpreter now does the sums.