SciCode Programming Problems: 22,532 AI-Generated Scientific Computing Tasks

22,532 programming problems generated by AI, inspired by real scientific computing code snippets. Each problem is paired with a solution and focuses on concepts like numerical algorithms, data analysis, and mathematical modeling. The dataset was created by SciCode and was last updated on 2026-02-19.

Use Cases

• Train or evaluate code generation models based on AI-generated problem-solution pairs.
• Benchmark AI performance on scientific computing tasks based on described concepts like numerical algorithms.
• Develop educational tools for programming based on a large corpus of generated problems.
• Study the characteristics of AI-generated code compared to human-written scientific code.

Strengths

• Large scale with 22,532 problem-solution pairs.
• Focus on scientific computing concepts as described, including numerical algorithms and data analysis.
• Each problem is paired with a comprehensive solution.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the AI generation process and source code snippets.

Provenance

Source

SciCode via Hugging Face.

Collection Method

AI-generated problems inspired by real scientific computing code snippets.

Freshness

Last updated 2026-02-19 03:24:58; freshness should be verified.