DeepSeek: Disrupting the AI Landscape with Innovation and Open Access

There is no question that the advent of Generative AI (Gen AI) has created a new paradigm that sent shockwaves across many industries. The tool in question is based on the so-called Generative Pre-trained Transformer (GPT) models – the models that generate human-like text with seemingly sophisticated reasoning. Most of the world is still coming to grips with the new technology with many academic institutions, including universities, opting to ban the use of GPTs altogether whilst industry is frantically hiring AI strategists to make sense of any new opportunities and threats the new technology brings. More encouragingly, many CxOs have issued ‘Gen AI first’ calls to their organisations.???

The most popular of the platforms is Chat GPT that was developed by the US-based research organisation OpenAI. OpenAI’s experience in developing the technology was seen by investors and other technology observers as the ‘ground truth’ providing insights on the overheads required to develop and run the technology. Consequently, Investment in supporting industries skyrocketed pushing, for example, GPU chipmaker Nvidia above the $1trillion cap – against the expectation that massive infrastructure is needed to develop and run the technology.??

In releasing their most recent model, Chinse-based Deepseek have demonstrated that Gen AI capabilities can be built and delivered at a fraction of the overhead expectations set by OpenAI. Reported development costs were around ￡5m for R1 AI model. By contrast, OpenAI's development costs for GPT 4 exceeded ￡80m, with both models achieving equivalent performance. Moreover, Deepseek have released their model to the wider community as open source – allowing developers to run versions of the model on their own infrastructures to minimise data privacy issues. Initial experiments we carried out show that the overheads of running Deepseek’s model locally seem to be significantly lower than other open-source models. Deepseek have followed the model that Chinese industry has been running for decades – provide products at scale and at costs that Western industries can never compete with. This is model is no further reinforced by the release of Qwen 2.5-Max by Alibaba which, reportedly, outperforms its peers – potentially kickstarting the commercialising of AI in China.??

Initial feedback from the AI world has been mixed. There is the issue of data privacy and IP which is a concern across most, if not all, Gen AI platforms. There is the issue on cultural biases – again, true across all platforms. There is no doubt that prolonged use of the new platform will unearth more observations. Notwithstanding, the AI race across opposing political ends is now on and will fuel even more rapid development in the field. This means that AI and data literacy for everyone will become a necessity to enable survival and growth of organisations.???

Training in Data is focused on providing organisations with state-of-the art training programmes that empower and enable every employee to take advantage of new technologies safely and securely. Our data literacy programmes are proven to deliver significant? gains, for individuals and organisations, via appropriate use of current and emerging AI technologies.???

?

Dr Hasan Al-Madfai,

CTO – Training in Data?