tooling · program-aided-computation TEC-0016

Offload arithmetic to code

Have the model write a short program for the numeric part and run it, instead of computing in text.

The model reads the problem and produces code that performs the calculation, then an interpreter executes it. The model does the decomposition it is good at and the interpreter does the arithmetic it is bad at. Works with any tool-calling setup that can run Python.

Requires: Requires a sandboxed interpreter.

Addresses: Digit-level arithmetic

Contexts: Math, Data analysis

Does it work?

argued, not measured1 supporting · 0 contesting source · last moved 2026-09-04

Efficacy claims — what this technique actually moves, under which conditions, and whether that has been contested.

Needs: an environment where code can run

Fails when:

The error is in translating the problem into a program, which is now the only thing the model does. Verification has to move there with it. (helps, narrowly)

mechanismmechanism reasoning
Offloading the calculation to generated code removes the arithmetic as a source of error but does not remove error — what remains is the translation from problem to program, which is now the only thing the model is doing. Verification has to move there with it.
- supports PAL: Program-aided Language Models (2022)

Code

https://github.com/reasoning-machines/pal — Reference implementation with prompts and evaluation on math word problems.verified 2026-09-02

Sources

PAL: Program-aided Language Models (2022)

Either way, bring evidence: a link, or what you saw.

Whether a technique works is a claim. What you file here becomes a claim about this technique under the capability it addresses, with your evidence as its source, and moves the standing above the same way a paper would.