Google AI Legend's Take on DeepSeek Shock: "The Power of Open Source" [Tech Talk]

'Attention' Paper Author Ilya Polosukhin
"Open Source AI That Anyone Can Contribute To"
"The Speed of Innovation Outpaces Large Corporations"

"DeepSeek was able to stay months ahead of OpenAI because it was open source."

This was the assessment made by Ilya Polosukhin, CEO of Near Protocol, during the 'Open Source AI Media Day' dialogue hosted by Vidl Asia on the 11th. Polosukhin, a startup entrepreneur formerly with Google AI and a scholar who played a crucial role in the birth of modern AI, predicted that open source would be the main driver of competition in AI development.

"The speed of innovation is the true strength of open source"

Ilya Polosukin, CEO of Near Protocol, speaking at the 'Open Source AI Media Day' panel held on the 11th at Lotte World Tower in Seoul. Photo by Juhyung Lim skepped@

'Open source' refers to an open collaborative development model. All developers can access the software (SW) source code without any cost or compensation, and anyone can add or modify the code. Open source AI applies this free and open development approach to AI. Many large companies, including Meta, and numerous global startups contribute to the open source AI ecosystem. DeepSeek, developed in China, is one of them.

Polosukhin CEO described DeepSeek as "a model that opened new possibilities in reinforcement learning, one of the AI models," but also explained, "In fact, OpenAI had been attempting a similar project for several months. However, in the end, their release was later than DeepSeek's." Why was the large company OpenAI trailing behind DeepSeek?

Polosukhin CEO emphasized, "Large companies research and test numerous products internally, but the final version must go through a long process. However, DeepSeek released its developing model as open source and was able to gradually improve it together with other contributors. This is the speed of innovation, the strength of open source."

'Attention' creator and Google AI legend... Why he became an open source evangelist

The CEO of Polosukin is one of the authors of the paper titled "All You Need Is Attention," in which the concept of "attention" introduced led to the development of the transformer, a core technology in advanced AI today. Arxiv capture

Polosukhin CEO is someone who understands the innovation speed of large companies better than anyone else. He majored in computer engineering at Kharkiv Polytechnic University in Ukraine and worked as a researcher at Google AI starting in 2014. This was when Google acquired the UK-based DeepMind and began focusing seriously on neural network AI research. After three years of research, Polosukhin CEO and the AI team published the paper "Attention is All You Need." This paper led to the 'Transformer,' which is the backbone of almost all advanced AI models today. Numerous models, including OpenAI's ChatGPT and Google's Gemini, run on the Transformer architecture.

After achieving this tremendous accomplishment, Polosukhin CEO founded the startup Near Protocol instead of staying at Google. The company’s mission is to expand the open source AI development ecosystem using blockchain and cryptocurrency technologies. Polosukhin CEO pointed out that major AI companies like Google and OpenAI are actually developing "closed models," saying, "Models promoted as open are closer to open parameters. The core data that make up the model are hidden, so neither developers nor consumers really know."

Instead, he emphasized, "Open source allows anyone to access the core source code and contribute to optimizing the model. Also, contributors are motivated by their own profit, which creates an incentive to advance the technology faster. Moreover, since it is not a product made by a company but a customized AI made by individuals to meet their own needs, it is more advantageous in terms of optimization."

Polosukhin CEO’s Near Protocol was established to further promote idea exchange and monetization among open source AI developers and consumers. It aims to make the payment system transparent using blockchain technology, which prevents ledger manipulation, and also to solve security issues, a vulnerability of open source.

"Not only Silicon Valley but everyone can share the benefits"

Silicon Valley, USA. Pixabay

Polosukhin CEO emphasized that mid-sized countries like South Korea can also enhance their competitiveness through open source AI. He explained, "The process of creating AI, including model development and data collection, requires significant computing resources. On the other hand, open source builds continuously on the foundation created by someone else. In the past, when new technologies were developed, only some companies in Silicon Valley benefited, but through open source, all companies and individuals can share the benefits."

He added, "South Korea is already a leader in applied technology. Also, open source AI can be integrated into industries where South Korea is already prominent, such as entertainment. There will be opportunities for startups to develop models trained on Korea’s content know-how and unique data, curated to suit global consumers."

However, open source AI still faces hurdles. Polosukhin CEO pointed out, "Advanced countries including South Korea have strict laws related to privacy and data collection, which may make it difficult for startups to access high-quality data."

Accessibility issues also remain a challenge compared to large company products. For example, OpenAI’s ChatGPT can be used simply by entering requests on a webpage or application, but most open source AI requires installing the model on a computer, which is complicated for the general public.

Polosukhin CEO said, "(Accessibility) is definitely a task that many startups need to develop," but emphasized, "Open source has the characteristic of 'customization.' Companies like OpenAI and Google, which must develop universal products for as many customers as possible, cannot fulfill this, making it a unique strength of open source."