Bridge (Remediating Math Mistakes)
Bridge is a research dataset released with the NAACL 2024 paper "Bridging the Novice-Expert Gap via Models of Decision-Making: A Case Study on Remediating Math Mistakes" by Rose E. Wang and colleagues at Stanford University. It gathers 700 real tutoring conversation snippets drawn from grade 2 to 8 mathematics sessions, each captured at the moment a student makes a mistake. Every snippet pairs the original novice tutor turn with an expert tutor's revised response, and adds three expert annotations that model the reasoning behind the revision: the error the student made, the remediation strategy chosen, and the intention behind it. The topics span place value, rounding, fractions and decimals, multiplication and division, word problems, geometry, measurement and early functions. You access it freely from Hugging Face (auto-converted Parquet, loadable with the datasets library or pandas) or from the GitHub repository, split into 419 training, 71 validation and 210 test examples. For RAG it is a small, clean, well-structured corpus of annotated pedagogy rather than subject content. For an AI tutor it is unusually specific: most sources here supply material to teach from, whereas Bridge supplies the expert decision-making for how to respond when a learner is wrong, which is directly useful for modelling remediation moves, coaching weaker responses and evaluating a tutor's pedagogy. The licence is Creative Commons Attribution-NonCommercial 4.0, so attribution is required and commercial use is prohibited: this matters for anyone building a paid tutor, because the conversations are real (anonymised) student and tutor turns that cannot be used commercially without separate permission. The repository code carries a separate permissive licence, but the dataset itself is non-commercial. It complements the maths and education sources we already list: where openstax, ck-12, khan-academy and siyavula give curriculum text, and amps, megamath, stackmathqa and naturalproofs give problems and proofs, Bridge sits alongside case-network as a record of tutoring behaviour, with common-core and ngss providing the standards its lessons map to.
Related sources
AMPS
A dataset of informal mathematics introduced alongside the MATH benchmark. It includes more than 100,000 Khan Academy problems with step-by-step solutions in LaTeX and over 5 million problems generated with Mathematica scripts, totalling around 23 GB.
AutoMathText
Around 200 GB of mathematical text compiled from websites, arXiv, and GitHub, drawing on OpenWebMath, RedPajama, and AlgebraicStack. Every piece of content carries a score from 0 to 1 for relevance, quality, and educational value, labelled automatically by the Qwen-72B model.
Awesome AI Math Datasets
A community-curated list of open-source mathematics datasets for training and evaluating maths-capable language models. A useful index for finding newer additions in this space.
Big-Math-RL-Verified
A verified maths bank of 251,122 problems, each filtered to a single checkable answer and tagged with a model solve-rate as an empirical difficulty signal. Built by SynthLabs and released under Apache 2.0.