Google's Minerva Will Solve a Third of Practice Problems in University Mathematics, Physics, Chemistry, Economics, and Biology!
We have another exciting development for language model enthusiasts. The Minerva model is based on Google's Pathways language model, which contains 540 billion parameters. It was trained on...
We have another exciting development for language model enthusiasts. The Minerva model is based on Google's Pathways language model, which contains 540 billion parameters. It was initially trained on general language data and then fine-tuned on technical content. It appears to be the best of its kind in its focus now. It addresses mathematical and scientific questions using a "step-by-step" reasoning approach. The model was trained on a 118GB dataset of scientific papers from arXiv and websites that feature mathematical expressions using LaTeX, MathJax, or other mathematical formats.
Sources:
- https://ai.googleblog.com/2022/06/minerva-solving-quantitative-reasoning.html?fbclid=IwAR3y99mIqQdDVeoDvlQsIMyOOmdEhABoVsFNerC8MmqfJfgX7fCXtLqG8vs
- Research paper on Minerva: https://arxiv.org/abs/2206.14858?fbclid=IwAR3xEJ3qFOFveZtEAUBxLqYDdlF711Dc0_qcUJonWbPtSHyGP3-9QZsGJ2o
- Chain of Thought explanation from Google: https://ai.googleblog.com/2022/05/language-models-perform-reasoning-via.html?fbclid=IwAR3vqUMZDMAfEzpyiFbDDtdJ7qNmQ-oGpJtuyZ55kwOqPIK5fFdDHTPeWlo
- Research on Chain of Thought: https://arxiv.org/abs/2201.11903?fbclid=IwAR3xEJ3qFOFveZtEAUBxLqYDdlF711Dc0_qcUJonWbPtSHyGP3-9QZsGJ2o
- Meier, RD, Cheney, DL, and Seyfarth, RM (1990). Do Animals Remember the Past and Plan for the Future? Behavioral and Brain Sciences, 13(1), 1–16.
- https://pub.towardsai.net/how-minerva-a-new-nlp-model-performs-better-than-gpt-3-f7713458b59b
Původní zdroj: wordpress