Why Language Models Get Math Wrong
Large language models impress with code and prose, but they still frequently get multiplication wrong. The reason is simple: an LLM typically predicts the next token rather than executing a strict arithmetic algorithm. That's why, with long numbers, carries, and uncommon examples, the model starts guessing instead of calculating precisely. External calculators, code, and specialized modules come to the rescue.
AI-processed from Habr AI; edited by Hamidun News
Large language models (text-generating neural networks) can write code, translate, and hold conversations, but they have a systemic weakness when it comes to calculation and multiplication.
Why does a neural network get multiplication wrong?
A neural network doesn't calculate step by step the way people do. Instead, it predicts the most likely sequence of characters. Simple arithmetic exposes this quickly.
How do people differ from neural networks when calculating?
For a person, multiplication is an algorithm: break the numbers into place values, multiply the parts, carry the tens, and add up the result. A neural network sees an expression like 37 × 48 as a text pattern among millions of other sequences.
Key Questions Answered
Why does a neural network get multiplication wrong?
A neural network doesn't calculate step by step the way people do. Instead, it predicts the most likely sequence of characters. Simple arithmetic exposes this quickly.
How do people differ from neural networks when calculating?
For a person, multiplication is an algorithm: break the numbers into place values, multiply the parts, carry the tens, and add up the result. A neural network sees an expression like 37 × 48 as a text pattern among millions of other sequences.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.