What happened
Researchers found that asking AI language models to first write out the governing equations before solving a problem — rather than jumping straight to step-by-step reasoning or code — improves accuracy by 5–13% across different types of math problems. This matters because it shows a simple change in how you phrase a question to an AI can unlock better performance on practical problems like financial calculations or physics, suggesting the model already 'knows' the equations but needs them explicitly named to use that knowledge reliably.
Why it matters
This is an AI research result showing a measurable improvement on academic benchmarks, but the improvement is largest in applied domains (finance, physics) rather than pure math — which hints that explicit equation-first reasoning might help AI actually work better on real-world problems people care about solving, though real-world deployment data is absent.