November 8, 2022
Download the file
Oops! Something went wrong while submitting the form.

Global AI Trend 08: How Will Hyperscale AI Shape the Future?

J.A.R.V.I.S., F.R.I.D.A.Y., Vision, Ultron, ... You should find these names familiar if you have watched any of the 30 Marvel films. But if you haven’t, these are the names of the hyperscale AIs that not only assist the heroes in the movies, but also materialize into a human form and act as an ACTUAL MEMBER of the Avengers. And guess what, these super-cpable hyperscale AIs are no longer limited to movies; They have already become part of our reality and helped humans live a comfortable life and work efficiently! Come follow us! We, Crowdworks, will show you the A to Z of hyperscale AI technology that giant tech companies like Google have pledged to bring into reality.


Understanding Hyperscale AI

What exactly is hyperscale AI, which seems to be the buzzword for almost all latest AI trends? Hyperscale AI is a type of AI that imitates the human brain, i.e. an AI with a great number of parameters. Given that AI’s parameters are equivalent to the human brain’s synapses which are responsible for learning and remembering information, the large number of parameters would signify a much more advanced AI.

And yes, this hyperscale AI is even smarter -hundreds times or even more- than AlphaGo which is one of the smartest AI known to mankind so far; No wonder it’s been loved and developed by Google, Microsoft, Naver, Kakao and other big tech companies!

Hyperscale AI as a Superstar of Giant Tech Companies

Various domestic and foreign big tech and ICT companies are scrambling to build hyperscale AI. This is because hyperscale AI demonstrates higher productivity with a speed multiple times faster than that of existing AI and hence more value added to any field the AI is applied to. Truly, hyperscale AI has proven to be helpful in a wide range of business operations from product development to product design. But for today, let’s zoom into the wonders hyperscale AI has done in the tech industry in particular!

Ever heard of LG's creative artist Tilda? Yes, it’s the one that won the gold medal at the New York Festival which is one of the world's top three advertising festivals. That Tilda was built with LG's hyperscale AI EXAONE. To enable the birth of Tilda, EXAONE had to be trained on 600 billion corpora and 250 million image-text pairs with as many as 300 billion parameters. As a result of the intensive training with data from LG Chemistry and LG U+, LG’s subsidiaries, EXAONE became the powerhouse for image conversion technology; It is now capable of converting image to text and vice versa.

2. Naver's HyperCLOVA
In May, Naver introduced the very first Korean-based hyperscale AI, HyperCLOVA. Since then, HyperCLOVA worked with extensive data from the blogs and news on the nationwide-popular platform Naver and successfully built a conversational AI. This conversational AI is “advanced interactive” in a sense that it can hold a very natural conversation on any topic just like you could with your human friends. Not only that, but Hyper Clova is also capable of text generation and conversion. It’s been said that Hyper Clova will become available to public institutes and universities for research purposes.

3. Google's PaLM
This year, Google unveiled its hyperscale AI PaLM, which outperforms traditional AI. Known as a language model that can handle programming and math problems, PaLM boasts the surprising 540 billion parameters. And word has it that it’s three times larger than GPT-3, a leading AI of the day. No wonder that PaLm can read between the lines as well as interpret the text based on its training on multilingual datasets and that its arithmetic skills are on a par with that of human beings. Human language, math, and… the list doesn’t end here. PaLM turns out to be excellent at coding as well while it uses Python codes 50 times less than the existing AI. What a master of all trades it is!

The Future of Korean Hyperscale AI

Resource-wise, Korean companies in the field of hyperscale AI lack researchers and data compared to their global competitors like Google and Microsoft; Strategy-wise, Korean companies would be better off cooperating and complementing each other than competing if they are to accomplish hyperscale AI of an advanced level. For sure, collective effort to advance and commercialize hyperscale AI based on camaraderie will bring Korea one step closer to a better future with AI.

AI For Everyone, Crowdworks

Other whitepapers you might like

Find out how reliable training data can give you the confidence to deploy AI

Get started