Solving Quantitative Reasoning Problems with Language Models
FreeLarge language model for solving quantitative reasoning problems across sciences and engineering
FreeFree tier
About Solving Quantitative Reasoning Problems with Language Models
Minerva is a large language model introduced by Google researchers in the paper 'Solving Quantitative Reasoning Problems with Language Models.' It is pretrained on general natural language data and further trained on technical content, enabling it to solve quantitative reasoning problems across mathematics, science, and engineering at the undergraduate level. The model achieves state-of-the-art performance on technical benchmarks without relying on external tools, correctly answering nearly a third of over two hundred college-level problems from physics, biology, chemistry, economics, and other sciences.
Key Features
Pretrained on general natural language data and further trained on technical content
Achieves state-of-the-art performance on technical benchmarks without external tools
Evaluated on over two hundred undergraduate-level problems in physics, biology, chemistry, economics, and other sciences
Capable of solving mathematics, science, and engineering problems requiring quantitative reasoning
Pros & Cons
Pros
- State-of-the-art on technical benchmarks without needing external tools
- Trained on a broad range of technical content for robust quantitative reasoning
- Demonstrates ability to solve complex problems across multiple scientific disciplines
Cons
- Correctly answers only about a third of undergraduate-level problems, indicating limited accuracy
- Struggles with tasks requiring quantitative reasoning, a known limitation of prior models
- Requires large-scale pretraining and specialized technical data, which may not be accessible to all researchers
Best For
Solving undergraduate-level quantitative reasoning problems in physics, biology, chemistry, economics, and other sciencesAnswering college-level mathematics and engineering problemsBenchmarking language model performance on technical reasoning tasks
FAQ
What is Minerva?
Minerva is a large language model pretrained on general natural language data and further trained on technical content to solve quantitative reasoning problems in science, engineering, and mathematics.
Does Minerva use external tools for reasoning?
No, Minerva achieves state-of-the-art performance on technical benchmarks without the use of external tools.
What types of problems can Minerva solve?
Minerva is evaluated on over two hundred undergraduate-level problems in physics, biology, chemistry, economics, and other sciences that require quantitative reasoning.
How accurate is Minerva on undergraduate-level problems?
Minerva can correctly answer nearly a third of the undergraduate-level problems it was tested on, indicating room for improvement.