Soon after US president Donald Trump; announced the Stargate Project offering financial support of $500 billion; towards AI research and development, the world of AI was hit by a bombshell with the announcement of DeepSeek AI model R1. It was designed and built in China by founder Liang Wenfeng, who holds a degree in both Electronic Engineering and Computer Science and is also the current CEO of a hedge fund called High-Flyer. This hedge fund uses AI to analyse financial data to make investment decisions, a process called quantitative trading.
The main reason DeepSeek shook the AI community was by having development costs a fraction of the cost of AI LLMs developed in the USA, despite an import embargo on AI chips that the US imposed on China. AI stocks took a nose-dive after the announcement with company’s involved in AI hardware and software losing as much as 20% of their market value.
Wenfeng was able to perform this miraculous task by stockpiling a huge volume of older generation GPUs over the course of the year and using these to build his AI engine. DeepSeek R1; was released to the public to little fanfare - compared to existing online LLMs – as an opensource AI engine on 20 January 2025.
Despite being open to the public, many world leaders are sceptical of its intentions, with many countries, including USA and Australia, banning its use on government devices and systems, citing a national security risk.
Another bone of contention came from a well-known AI company, OpenAI, whose ChatGPT app fell to second place on the app store behind DeepSeek AI assistant within days of its release. CEO of OpenAI, Sam Altman, has accused DeepSeek of using OpenAI’s search results to generate its own responses. In the AI world this is known as ‘knowledge distillation’ or simply ‘distilling’ and is a technique where a large and complex AI model, known as a teacher, transfers its knowledge to a smaller, more efficient student learning model. This allows the student model to perform similar tasks to its bigger brother, while being faster and requiring less computational power [read energy].
Altman has said in a statement that “they believe that a Chinese startup called DeepSeek has used proprietary data from ChatGPT to train their own AI model”. He is essentially accusing them of intellectual property theft. I find the irony of this accusation quite humorous as ChatGPT was initially accused of using proprietary information on the web in its own training, a process that Altman has dismissed as being above board.
However this story may pan out, I believe that the introduction of DeepSeek R1; underscores an industry-wide shift towards more cost-effective AI infrastructure.
| Tel: | +27 11 543 5800 |
| Email: | [email protected] |
| www: | www.technews.co.za |
| Articles: | More information and articles about Technews Publishing |
© Technews Publishing (Pty) Ltd | All Rights Reserved
printer friendly version