Generative AI and New Geopolitics

Why does the US government restrict companies like NVIDIA from supplying GPUs to other countries, including China? What did the fall in NVIDIA and other tech giants of the USA share prices with the launch of the DeepSeek signal? What is the response of the USA and other countries?

- Prof. (Dr.) Nityesh Bhatt and Prof. Anand Kumar

The term “Artificial Intelligence (AI)” was first coined by John McCarthy in 1956. However, the discussion of concepts that shaped research on AI began in the 1940s. Today, we have reached the frontier of Gen AI on this path. It is an interactive system that creates new content like conversations, stories, images, videos, music, etc, based on the types of inputs given (prompt engineering).

ChatGPT was the first Gen AI model launched by OpenAI (founded in December 2015) on Nov. 30, 2022. Within five days, the chatbot had attracted over one million users. After the launch of ChatGPT, many companies strategically developed their LLMs to stay relevant to the market. Some of the most notable ones from the USA include Bard, later changed to Gemini (by Google), Claude (by Anthropic), and LLaMA (by Meta), Co-Pilot by Microsoft, Perplexity AI, Amazon CodeWhisperer, Mistral AI from France, Qwen by Alibaba from China, etc. Another interesting development is Devin AI that is an autonomous AI assistant tool created by Cognition Labs that can write, debug, deploy software applications and manage IT projects on its own.

On Jan. 20, 2025, DeepSeek Gen AI, based on an open-source technology (released under MIT license) and owned by Chinese hedge fund High-Flyer, was launched. Coming from a start-up launched in 2023, DeepSeek followed multiple innovations like reinforcement learning, reward engineering, distillations, and emergent behaviour networks. With a small budget of $5.6 million to train its model with 671 billion parameters, the DeepSeek-R1 model was considered to have similar capabilities as OpenAI's GPT or model, which cost around $600 million (similarly US$500 million of Meta's Llama 3.1) and an estimated 3,500 people with two years' time duration.