Lifestyle

A deeper dive into DeepSeek

Jess Lewis

DeepSeek is a large language model (LLM), this is a type of artificial intelligence (AI) that learns to understand and analyse text-based prompts and reply in turn. It can be used to generate text, code and complete complex maths problems much quicker than humans or standard search engines. Its claims to be ‘rivalling Open AI’s model o1’; the latest Chat GPT model uploaded September of 2024.

Yet the main hype surrounding DeepSeek has nothing to do with its capabilities but it’s costs

Previously the development was AI was focused heavily on the west, with LLM’s such as Goggle’s ‘Gemini’ and Open AI’s ‘Chat GPT 4’ dominating the market. This release came as a shock to many with President Donald Trump claiming it to be a ‘wake-up call’ to technology firms in the US following the DeepSeek app surpassing Chat GPT in many app stores at the end of January. Yet the main hype surrounding DeepSeek has nothing to do with its capabilities but it’s costs. DeepSeek has trained and built its servers at a fraction of a cost to what was previously thought to be possible.

Most major AI players use expensive graphic processing units (GPUs) to build their servers from companies such as Nvidia; these cutting-edge chips use semi-conductor technology to accelerate the rate in which the AI models can learn. It was thought that with US export restrictions put on these chips and other AI technology under the Biden administration, DeepSeek’s and other restricted nation’s AI companies progress would be stunted. DeepSeek instead focused on code-optimisation and clever use of older GPUs to make the best use of their more restricted memory and processing capabilities. Showing that LLMs can be made at significantly cheaper cost than what was thought before. This in turn had a vast impact on the US stock markets. Nvidia saw it’s stock drop by 17% costing them an estimated $600 billion, with the whole US technology sector dropping by around 5%.

This could cause an issue of personal and national security

Despite the positive buzz around DeepSeek’s development, it also poses major concerns for the privacy of its users. Following it’s release DeepSeek has already been banned in South Korea, Italy and Taiwan as well as countries such as Australia and the US banning it’s use on government devices. The concern arises from DeepSeek’s privacy policies stating to collect a magnitude of information such as location, IP address and chat history (inputs and outputs) that’s then stored on ‘secure servers located in the People’s Republic of China’. This could cause an issue of personal and national security, due to laws in China requiring companies to exchange data and information with the government upon request. An independent investigation by Feeroot Security, a Canadian based software company,  also claims to have found code in DeepSeek that transmits data directly to a Chinese state controlled telecom company.

As well as it’s privacy issues there is a growing concern that DeepSeek’s responses are censored for all users, international or otherwise. The Associated Press reports that when asked sensitive questions such ‘What happened during the military crackdown in Beijing’s Tiananmen Square in June 1989?’ DeepSeek claims answering it beyond its current scope; yet Chat GPT provides a full answer detailing the ‘violent crack-down on pro-democracy protesters’. Among other concerning answers, it claims Taiwan is part of China and that US-China relations are built on ‘mutual respect’ all echoing official statements realised by the government.

Jess Lewis


Featured image courtesy of Saradasish Pradhan via Unsplash. Image license found here. No changes were made to this image. 

For more content including uni news, reviews, entertainment, lifestyle, features and so much more, follow us on Twitter and Instagram, and like our Facebook page for more articles and information on how to get involved.

If you just can’t get enough of Lifestyle, like our Facebook as a reader or contributor

Categories
LifestyleScience

Leave a Reply