New Datasets and Models Accelerate AI

It seems that the miraculous age of artificial intelligence is continuously accelerating. Every day, I come across several intriguing ways to speed up, enhance, or refine artificial brains. New fantastic datasets and models are emerging.

For instance, we have a top-notch model for generating free text, GPT-J-6B, which outperforms GPT-Neo. Two enthusiasts trained a beautiful model with 6 billion parameters on a single TPU in just 5 weeks.

A fast, accurate, and accessible programme has emerged that translates between Czech and English significantly better than, say, Google Translator, and it’s called DeepL.

And finally, perhaps most interestingly, on 1st June, the Beijing Academy of Artificial Intelligence (BAAI) introduced the Wu Dao 2.0 model (Chinese for enlightenment). This network is the largest neural network ever created and likely the most powerful. It improves upon the original Wu Dao, has only been with us for a few days, and already legends about it are circulating on the internet.

If the 19th century was the age of steam and the 20th century the age of electricity, then the 21st century is the age of computers, information, the internet, and artificial intelligence. Or as AI wrote while helping me draft this article: The age of artificial intelligence is nothing less than a revolution!

Sources:

• https://arankomatsuzaki.wordpress.com/2021/06/04/gpt-j/

• https://www.deepl.com/translator

• https://towardsdatascience.com/gpt-3-scared-you-meet-wu…

+ an interesting bonus until 17th June: https://www.ceskatelevize.cz/…/421235100121014-clovek…/