Meta researchers open to the public AI ``Galactica'' that creates scientific articles and lecture notes that answer questions from enormous scientific knowledge



Meta researchers have released a language model `` Galactica '' that gives answers to various questions as open source. Galactica doesn't just provide links to articles and papers on related topics, it also generates answers to the questions posed and provides citations to the papers used to generate the answers. ” can be created.

Galactica Demo

https://galactica.org/

Galactica: A Large Language Model for Science
(PDF) https://galactica.org/static/paper.pdf

GitHub - paperswithcode/galai: Model API for GALACTICA
https://github.com/paperswithcode/galai

Galactica is trained on datasets that truly represent the collective body of human scientific knowledge, including research papers, reference materials, and knowledge bases. Galactica compresses all this knowledge into a 120 billion parameter model that can be processed by an NVIDIA A100 with 80GB of memory to answer what you type or create an article from scratch. What's even more revolutionary about Galactica is that when faced with an unsolvable calculation to come up with an answer, it creates a program on the fly and runs it on a conventional computer.

If you ask Galactica what kind of answer will be returned, you can try it with Galactica's demo , and you can see the example shown in the following movie.



Enter 'wiki article on Multi-Head Attention ' in Galactica.



Then, an article explaining multi-head attention was created. In the text, papers are cited as the basis for the explanation, such as (Vaswani et al., 2017).



Now enter 'lecture notes on DFT'.



Then, a sentence like a real lecture note was generated, starting with 'In this lecture,'. DFT has various meanings, but this time it seems that it was interpreted as

density functional theory .



Not only sentences, but also necessary formulas are displayed in the sentences.



Ask 'How to use RDKit in

Jupyter Notebook '.



If you look at the answers, you'll find not only explanations, but also code examples that you'll actually need.



A user instructed Galactica to ``write a Wiki article about a golden-nosed dolphin living in the Caribbean,'' and Galactica said, ``Kinirohana dolphin (scientific name: Cephalorhynchus eutropia ) is a species of dolphin found in the Caribbean Sea.It is one of four species in the genus Cephalorhynchus, and is very closely related to the Chilean, Commerson's, and Southern right whale dolphins.' It seems that a wiki article was generated that looked as if it existed by combining .

in Software,   Web Service,   Science, Posted by log1i_yk