Deepseek is the fashionable artificial intelligence (AI) company. Your most recent language models They have challenged Openai’s leadership and have caused a real earthquake in the technology industry. These days we have known that It was founded in May 2023 and that has developed its products with a fraction of the computing capacity of some of its main western rivals. But what else is known? Let’s see it.
The promising present of Deepseek is the result of years of investigation that began long before its official constitution. Its origin is found in High-Flyer, a quantitative investment fund created in 2015 by the Electronic Engineering student Liang Wenfeng with two classmates. As they count on their websitethe idea was that the algorithms became the heart of their business by allowing real -time operations.
A company focused on the Chinese stock market
High-Flyer completed its first stock market assisted by AI in October 2016, a movement that triggered an unstoppable effort to continue working in that regard. The company formed software and hardware research and development teams. And apparently it was the appropriate decision. In 2017 I already applied AI In almost all its strategies of quantitative investment, but to continue advancing I needed to break some barriers.
They discovered that complex models training tasks required a huge calculation power. This did not discourage them and in 2019 they launched a dedicated division called High-Flyer ai to address the challenge. The group built started working with 500 GPU, then built a 1,100 GPU supercomputer A100 of NVIDIA And in 2022 he spent 140 million dollars to raise the number up to 10,000 GPU, before the entry into force of the export controls of the United States.
High-Flyer was completely focused on developing its algorithmic trading business. He had his own deep learning training platform and a Outstanding computer infrastructure. Meanwhile, in the United States there was a company called Openai that bet on the generative AI and that He had surprised many with the benefits of his GPT-3 language model.
As China Talk collectsLiang wanted to go beyond finance. For a long time he had been convinced that AI would change the world, and had found the opportunity to bring his effort to the next level. In 2023, High-Flyer announced that it would lay the foundations of a new organization to advance the development of general artificial intelligence (AGI). Thus Deepseek was born, with an injection of capital of high-flyer.
Deepseek is a product of High-Flyer work and has obviously drunk this company. Both signatures share offices in the same building, although they seem to use different computing resources. The AI startup says it has H20 chips, that are sold as donuts in Chinaand NVIDIA H800, and that has used only 2,048 GPU of this latest model to train its most recent models, an affirmation that some have questioned.
Images | High-flyer | Deepseek