ComputeGPT is a computational chat model that provides real-time accurate answers to numerical problems. The chat model is able to answer SAT questions, GRE questions, all kinds of math and science questions, and homework problems with high accuracy.
Note: This model also has other names, such as MathGPT and ScienceGPT, but we prefer ComputeGPT as it can compute anything, not restricted to certain domains. It can solve medical dosage problems, physics problems, typical CS problems, and more. Just compute your answer.
Just head over to https://computegpt.org and use the model to your desire! There are no limits on its use.
Feel free to come back here and report any issues or leave any feedback to help improve the model.
ComputeGPT is also available on mobile platforms, like Google Play and Apple App Store.
ComputeGPT is a free and accurate chat model and calculator for math, science, and engineering. It's also known as MathGPT and ScienceGPT, and can compute most numerical answers.
ComputeGPT is publicly available at no cost and can solve most textbook problems easily. It can handle advanced topics like Linear Algebra, Calculus, Physics, and more! Solve math with GPT!
Download, chat, and compute any answer!
We provide an evaluation and comparison of ComputeGPT on numerical problems against GPT-4 with Internet (Bing AI), GPT-3.5 (ChatGPT), GPT-3 (Davinci-003), and Wolfram Alpha Natural Language. See the full evaluation at https://github.com/ryanhlewis/ComputeGPTEval.
| Model | ComputeGPT | Wolfram Alpha | Davinci-003 | ChatGPT | GPT-4 |
|---|---|---|---|---|---|
| Overall Accuracy (%) | 98% | 56% | 28% | 48% | 64% |
| Word Problems (%) | 95% | 15% | 35% | 50% | 65% |
| Straightforward (%) | 100% | 83.3% | 23.3% | 46.6% | 63.3% |
Overall, ComputeGPT demonstrates state-of-the-art performance on numerical problems when compared against other models, on both straightforward problems and word problems that require more reasoning.
While most code models can handle logic, like "Is 5 < 3?", it will not be able to deduce a logical answer from a multiple-choice answer-set. It works best with open-ended numerical answers. In a choice scenario, append the logical answer as a question to the end of your prompt to get a true-false.
The source code is currently unavailable, but will be available soon will full licensing and open-source support. It is currently under heavy development by the Urban Information Lab.
If you would like to contribute to the Compute GPT codebase, have suggestions, or would like to provide unit tests, please open an Issue and let us know! It's very helpful to see what types of problems the model fails on in order to fine-tune it. Together, we can make a better computational GPT model.
Alternatively, if you have a private request, please email me at rhl@utexas.edu.
ComputeGPT: A computational chat model for numerical problems -- https://arxiv.org/abs/2305.06223
@article{lewis2023computegpt,
title={ComputeGPT: A computational chat model for numerical problems},
author={Ryan Hardesty Lewis and Junfeng Jiao},
journal={arXiv preprint arXiv:2305.06223},
year={2023},
}