Why AI tutors get maths wrong, and what to do about it
AI tools that write essays well still make errors in fairly basic arithmetic. That combination confuses people, and understanding why explains a lot about how to use them safely.
They predict text, they do not calculate
A language model is not doing arithmetic when it answers a sum. It is producing the text most likely to follow the question, based on patterns learned from an enormous amount of writing.
For common sums this works, because the correct answer appears constantly in the training data. For unusual numbers, multi-step problems or unusual phrasings, it is producing something that looks like a plausible answer rather than a computed one.
Why the errors are dangerous rather than merely wrong
- They are stated with complete confidence. There is no hesitation to alert you.
- The working can look right while the answer is wrong, or the answer right with invalid working.
- Errors cluster in multi-step problems, exactly where a child is least able to check.
- They are inconsistent. The same question can get different answers, which undermines a child who is trying to learn a method.
For a child, a confidently wrong answer is worse than no answer. They cannot evaluate it, so they learn it, and unlearning a wrongly learned method is much harder than learning it correctly first time.
Where it goes wrong most
Long multi-step word problems, larger numbers, unusual units, fractions and decimals combined, and anything needing exact rather than approximate answers. Also anything where the phrasing is unusual, since it is pattern matching against familiar phrasings.
How to use AI for maths safely
- Use it for explanation rather than for answers. "Why does this method work" is where it is genuinely strong.
- Check every numerical answer that matters. Especially anything a child will memorise.
- Prefer tools that draw practice from verified question banks over ones generating questions on demand.
- Ask it to show the steps, then check the steps rather than the answer.
- Teach your child that it can be wrong. A child who knows to check is safer than any filter.
What Different Drums does about it
This is precisely why the question bank exists. For arithmetic, maths, calculus, chemistry and physics, worksheet questions and answers are drawn from 200,000+ hand-verified items rather than generated when you press the button. Every one has been read and checked by a person.
It is slow, unglamorous work and it is the reason the maths can be trusted in a way generated maths cannot. The tutor still explains conversationally, because explanation is what models are actually good at, but the practice you print and the answers you mark against come from verified material.
Common questions
Are newer AI models better at maths?
Meaningfully better, yes, particularly the larger reasoning models. Better is not the same as reliable though, and errors still occur in exactly the places that matter for a learner. The gap narrows, it has not closed.
Can I just tell it to check its answer?
It helps somewhat and it is not a fix. A model asked to check its own work will often confirm its original answer confidently, including when that answer was wrong.
Learning that flexes around your child
Different Drums builds curricula, worksheets, tests and learning records around almost any topic, shaped to your child's pace, interests and your family's values. Core practice in maths and science comes from 200,000+ hand-verified questions. Ages 2-16+, used by families worldwide.
No credit card required. There is also a free guest mode with no sign-up.